Mixed-Criticality Scheduling on Multiprocessors using Task Grouping

Size: px
Start display at page:

Download "Mixed-Criticality Scheduling on Multiprocessors using Task Grouping"

Transcription

1 Mxed-Crtcalty Schedulng on Multprocessors usng Task Groupng Jankang Ren Lnh Th Xuan Phan School of Software Technology, Dalan Unversty of Technology, Chna Computer and Informaton Scence Department, Unversty of Pennsylvana, U.S.A. Abstract Real-tme systems are ncreasngly runnng a mx of tasks wth dfferent crtcalty levels: for nstance, unmanned aeral vehcle has multple software functons wth dfferent safety crtcalty levels, but runs them on a sngle, shared computatonal platform. In addton, these systems are ncreasngly deployed on multprocessor platforms because ths can help to reduce ther cost, space, weght, and power consumpton. To assure the safety of such systems, several mxed-crtcalty schedulng algorthms have been developed that can provde mxed-crtcalty tmng guarantees. However, most exstng algorthms have two mportant lmtatons: they do not guarantee strong solaton among the hgh-crtcalty tasks, and they offer poor real-tme performance for the low-crtcalty tasks. In ths paper, we present a parttoned schedulng scheme for mxed-crtcalty tasks on multprocessor platforms that addresses both ssues. Our schedulng scheme conssts of () a task-toprocessor packng algorthm that takes nto account the demands of tasks wth respect to ther crtcalty levels, and () a mxedcrtcalty unprocessor schedulng strategy that s based on task groupng. Our strategy assocates each hgh-crtcalty task wth a subset of the low-crtcalty tasks and encapsulates them n a task group, whch s scheduled wth the other task groups under the Earlest Deadlne Frst (EDF) polcy. Wthn each task group, the low-crtcalty task and the hgh-crtcalty tasks are scheduled usng a server-based strategy, so as to enable more of the former to meet ther deadlnes wthout affectng the latter. We present a schedulablty analyss for our schedulng strategy, and we show how tasks can be grouped usng Mxed Integer Nonlnear Programmng. Our evaluaton shows that our proposed scheme sgnfcantly outperforms exstng parttoned mxed-crtcalty schedulng algorthms, n terms of both the fracton of schedulable task sets and ts ablty to schedule low-crtcalty tasks. I. INTRODUCTION Due to recent advances n chp technology, multprocessor platforms are ncreasngly appled n safety-crtcal domans (such as avoncs, automotve, and ndustral control) to accommodate the ncreasng demand for computatonallyntensve workloads. Many systems n these domans are mxed-crtcalty (MC) real-tme systems that s, they ntegrate multple functonaltes (tasks) wth dfferent safety crtcalty levels, such as flght-crtcal and msson-crtcal tasks, on a sngle, shared hardware platform. Ths s typcally done to meet strngent requrements on space, weght, and power consumpton. Therefore, t s mportant to provde strong solaton and tmng guarantees for the mxed-crtcalty tasks wth respect to ther crtcaltes. Snce MC real-tme systems run a mx of tasks wth dfferent degrees of crtcalty, these systems are more complex and unpredctable than tradtonal real-tme systems. In order to certfy the correctness of a MC system, t s common to make certan assumptons about the worst-case behavor of the system for nstance, some task parameters are allowed to depend on the crtcalty levels of the tasks. A typcal approach s to assume more pessmstc worst-case executon tmes (WCETs) for hgher crtcalty levels for each task; n systems wth two levels of crtcalty, a task s assumed to be n the hgh-crtcalty mode f ts executon tme exceeds the low-crtcalty WCET. A maor challenge n ths approach s to guarantee mxed-crtcalty schedulablty (.e., hgh-crtcalty tasks always meet ther deadlnes and low-crtcalty tasks meet ther deadlnes f the hgh-crtcalty tasks are always n the low-crtcalty mode) of the tasks whle achevng effcent resource use. There s a rch lterature on schedulng mxed-crtcalty real-tme systems on both unprocessor and multprocessor platforms (e.g., [] []); however, most exstng solutons have at least one of the followng three key lmtatons. Frst, they assume that, whenever a ob of a hgh-crtcalty task exhbts hgh-crtcalty behavor that s, ts executon tme exceeds the task s WCET as estmated at the low-crtcalty level all current and future obs of every hgh-crtcalty task n the system wll also exhbt hgh-crtcalty behavor, whch does not always happen n general. Second, exstng technques cannot fully capture the dynamc behavor of the system: each task can dynamcally transton back and forth between the hgh-crtcalty and the low-crtcalty mode. Thrd, exstng solutons do not offer good performance for the low-crtcalty tasks: snce these are klled as soon as a hgh-crtcalty ob n the system enters ts hgh-crtcalty mode, t s possble that no low-crtcalty ob can ever meet ts deadlne for nstance, when the crtcalty change happens shortly after the system begns ts executon. Ths s undesrable because low-crtcalty tasks requre a certan tmng performance as well (although wth weaker guarantees than the hgh-crtcalty tasks) []. In ths paper, we propose a compostonal approach to mxed-crtcalty task schedulng that can solve these problems. Our approach, called TG-PEDF, provdes strong solaton among the hgh-crtcalty tasks, and t enables more lowcrtcalty tasks to be scheduled whle stll ensurng the mxedcrtcalty schedulablty guarantee of the entre system. TG- PEDF s based on task groupng and compostonal schedulng: each hgh-crtcalty task s encapsulated n a separate task group (to enable solaton) together wth a subset of the lowcrtcalty tasks (to enable overbookng of resources). Wthn each task group, the tasks are scheduled wth a server-based mxed-crtcalty schedulng strategy, and the dfferent task groups are scheduled on the platform as conventonal tasks under the EDF schedulng polcy. Ths ensures that the hghcrtcalty behavor of a hgh-crtcalty task cannot affect other hgh-crtcalty tasks, and t also enables more low-crtcalty obs n the system to meet ther deadlnes. Ths s because () the nterference of a hgh-crtcalty task on low-crtcalty tasks beyond ts local task group s sgnfcantly reduced, and because () low-crtcalty tasks are not always dscarded, but rather scheduled accordng to best-effort servce when the hgh-crtcalty task n the same task group s n the hghcrtcalty mode. Furthermore, our schedulng algorthm can dynamcally adapt to the crtcalty mode of each ob (by

2 enablng tasks to return to the low-crtcalty mode, dependng on the executon tme of ther current obs), and can thus save resources at runtme. Contrbutons. Ths paper makes the followng contrbutons: A parttoned mxed-crtcalty schedulng approach based on task groupng for mplct-deadlne perodc MC tasks on multprocessor platforms. In contrast to exstng solutons, our approach provdes strong solaton between hgh-crtcalty tasks and better tmng performance for low-crtcalty tasks; A schedulablty analyss for a set of mxed-crtcalty tasks that have been assgned to a sngle processor under our task group based schedulng algorthm; A Mxed Integer Nonlnear Programmng (MINLP) formulaton for the task groupng, based on the schedulablty constrants, that can mprove the schedulablty of the system; and A task packng algorthm that can save resources by takng nto account the demand of both low-crtcalty tasks and hgh-crtcalty tasks durng deployment. Our evaluaton on randomly generated task systems shows that the schedulablty of our proposed scheme outperforms the state-of-the-art MC schedulng algorthms. The rest of the paper s organzed as follows. In Secton II, we summarze the related work. Secton III presents the system model. The detaled descrpton of the task group based MC schedulng on a unprocessor s gven n Secton IV. Secton V presents the task packng algorthm. We dscuss our evaluaton n Secton VI and conclude the paper n Secton VII. II. RELATED WORK Snce Vestal s ntal work on mxed-crtcalty systems [], a rch lterature on mxed-crtcalty schedulng of real-tme systems has developed; see [3] for a survey. Tradtonally, work n ths doman has focused on unprocessor platforms [] [7], [4] [], but recently efforts have shfted towards multprocessor platforms. Several multprocessor mxed-crtcalty schedulng algorthms have already been developed [8] [], [] [], ncludng both global and parttoned approaches. For nstance, Pathan [] derved a suffcent schedulablty condton of a global fxed-prorty schedulng algorthm on preemptve multprocessors based on response tme analyss. Baruah et al. [] proposed global and parttoned schedulng algorthms for mxed-crtcalty mplct-deadlne sporadc task systems by combnng fpedf [3] wth EDF-VD, and they showed that parttoned schedulng gves better system schedulablty than global schedulng. For harmonc task systems, Mollson et al. [8] proposed an archtecture to use approprate schedulng technques for tasks wth dfferent crtcaltes on dfferent processors, whch provdes tmng solaton between these levels through a bandwdth reservaton server. Kelly et al. [9] studed the schedulablty of varous parttonng technques nspred by bn packng, as well as tradtonal parttonng heurstcs under fxed-prorty schedulng for MC systems. To address the effects of the physcal envronment on the system, Nz et al. [] presented a parttoned multprocessor schedulng scheme based on a mult-mode extenson of the zero slack rate-monotonc schedulng. Gu et al. [4] extended the demand-based sngle-processor mxed-crtcalty schedulng [] (called EY n ths paper) to multprocessor platforms and proposed two enhancements based on heavy low-crtcalty task awareness and balance factor to mprove the system schedulablty. However, these exstng solutons assume that, whenever a ob of a hgh-crtcalty task exhbts hgh-crtcalty behavor, all current and future obs of every hgh-crtcalty task n the system wll exhbt hgh-crtcalty behavor and thus all low-crtcalty tasks wll be dropped. Our work removes ths assumpton by consderng the specfc mode of each hgh-crtcalty tasks n the schedulng. There has also been work on offerng degraded servce to low-crtcalty tasks when the system s n the hgh-crtcalty mode. For nstance, Su et al. [7], [] studed an elastc mxed-crtcalty task model that allows low-crtcalty task to have dfferent perods n dfferent system modes. To maxmze the executon rate of low-crtcalty tasks, Jan et al. [8] ntroduced stretchng factors and proposed an assocated onlne decson algorthm based on the exstng elastc task model. Santy et al. [] proposed a technque that ams to avod or delay the rse n crtcalty level n order to mnmze the number of tasks that wll mss ther deadlnes. From the mplementaton perspectve, Burns et al. [9] ponted out the lmtatons of the exstng standard mxed-crtcalty models, and they proposed a more practcal model that consders the servce adapton of low-crtcalty tasks and the system mode swathng problem from the low-crtcalty mode to the hgh-crtcalty mode. Huang et al. [], [7] studed the servce reconfguraton and resettng for low-crtcalty tasks by quantfyng the degraded servce of low-crtcalty tasks. Gu et al. [] proposed a herarchcal executon model that can solate the mpact of hgh-crtcalty tasks from low-crtcalty tasks, and that can allow low-crtcalty executons n the hgh-crtcalty behavor by usng component boundares. Our work dffers from these approaches n that t can reduce the propagaton of hghcrtcalty ob behavor for mxed-crtcalty systems through the solaton of dfferent task groups. Further, our approach enables the system to swtch back to low-crtcalty mode from hgh-crtcalty mode once all hgh-crtcalty tasks exhbt lowcrtcalty ob behavor. To the best of our knowledge, the propagaton reducton of hgh-crtcalty ob behavor has not been nvestgated n exstng MC schedulng solutons. III. MIXED-CRITICALITY SYSTEM MODEL The system conssts of a set τ of mplct-deadlne perodc mxed-crtcalty tasks that need to be scheduled on a multprocessor platform wth p dentcal, unt-capacty processors. We follow the same task model as n []: Each task τ s characterzed by a tuple τ = (ζ, C τ (), C τ (), T τ ), where ζ {, } denotes the crtcalty level of τ ; C τ () and C τ () specfy the WCET estmates of τ at the low and hgh crtcalty levels, respectvely, where C τ () = C τ () f ζ =, and C τ () C τ () f ζ = (that s, the WCET at the hgh crtcalty level s more pessmstc than the WCET at the low crtcalty level); and T τ denotes the perod (relatve deadlne) of τ. We refer to C τ () and C τ () as the -WCET and - WCET of τ, respectvely. We assume that all tasks are ntally released smultaneously (at tme t = ).

3 Mxed-crtcalty schedulablty. A task set τ s consdered mxed-crtcalty schedulable under a schedulng polcy f the followng condtons are met: Low-crtcalty guarantee: If all obs run for at most ther -WCETs, all hgh-crtcalty and lowcrtcalty obs must complete before ther deadlnes. Hgh-crtcalty guarantee: If all hgh-crtcalty obs run for at most ther -WCETs and at least one runs for more than ts -WCET, all hgh-crtcalty obs must complete before ther deadlnes (whereas lowcrtcalty obs may be dropped). We note that although our system model and algorthms consder only two levels of crtcalty, t should be possble to extend them to more than two crtcalty levels. One potentally effcent approach s to perform task groupng n a herarchcal fashon accordng to the crtcalty levels of tasks, whch we plan to explore n future work. In the next secton, we present the algorthm for schedulng mxed-crtcalty tasks on a (un)processor. The algorthm for assgnng tasks to processors s dscussed n Secton V. Due to space constrants, we omt the proofs of lemmas and theorems here; they can be found n our techncal report [8]. IV. TASK-GROUP-BASED MIXED-CRITICALITY SCHEDULING ON UNIPROCESSORS To provde solaton between hgh-crtcalty tasks and to reduce ther nterference wth low-crtcalty tasks, we assocate each hgh-crtcalty task wth a subset of the low-crtcalty tasks and encapsulate both n a separate task group. Tasks wthn each task group are scheduled usng a server-based strategy to maxmze mxed-crtcalty schedulablty, and the task groups themselves are scheduled on the processor under the Earlest Deadlne Frst (EDF) polcy. (In a specal case where all tasks assgned to a processor have the same crtcalty level, they are scheduled as conventonal tasks usng EDF.) A. Task groups A task group conssts of one hgh-crtcalty task and several low-crtcalty tasks. It s specfed as T G = { τ, τ,..., τn, τ }, where the τ ( n) are the n low-crtcalty tasks, and τ s the sngle hgh-crtcalty task. The perod of a task group T G s a common dvsor of the perods of all tasks n the task group,.e., T T G = CD ( T τ, ) T τ,..., T τ n, T τ, where Tτ ( n) s the perod of the low-crtcalty task τ and T τ s the perod of the hghcrtcalty task τ. (Snce t s common for real-tme systems to have harmonc perods n practce, we make ths choce to smplfy the analyss; however, extensons to other perod values are possble.) All task groups are smultaneously actvated at tme t =. By defnton, the number of task group budget replenshments durng a perod of a low-crtcalty task τ s gven by l = T τ /T T G ( n), and the number of task group budget replenshments durng a perod of the hgh-crtcalty task τ s gven by h = T τ /T T G. We denote by B the mnmum budget (to be computed) that a task group T G must receve n each of ts perods to ensure all of ts tasks meet the mxed-crtcalty schedulablty. In our schedulng strategy, each task group T G wll be scheduled wth other task groups n the system as a conventonal perodc task wth perod (relatve deadlne) T T G and executon tme B. As usual, we defne the utlzaton of a task group to be ts budget dvded by ts perod. For ease of presentaton, we refer to a perod of a task group as a TG-perod. B. Three-phase Schedulng Strategy wthn Task Groups Basc deas. The schedulng wthn each task group T G s done n rounds; each round corresponds to a perod of the hgh-crtcalty task τ. Wthn each round, there can be three dfferent schedulng behavors: one s used whle the actual executon tme of τ s current ob has not yet reached ts -WCET; another s used when τ s ob mght reach ts -WCET durng the current perod of the task group; and a thrd s used when τ s ob has ether fnshed or exceeded ts -WCET,.e., swtched to hgh-crtcalty ob mode. Ths corresponds to three dstnct phases, whch are llustrated n Fgure : t May swtch to hgh-crtcalty ob mode Phase I Phase II Phase III Perod of task group Release of the hgh-crtcalty Absolute deadlne Fg. : A schedulng round of a task group. t + T Phase I (low-crtcalty mode): In ths phase, the current ob of τ has not yet reached ts -WCET and s thus n lowcrtcalty mode. All obs n the task group are scheduled to ensure the low-crtcalty guarantee. Phase II (swtchng mode): In ths phase, the current ob of τ can potentally exceed ts -WCET and thus swtch to hgh-crtcalty mode durng the current perod of the task group. All low-crtcalty tasks n the task group are suspended to ensure the hgh-crtcalty guarantee. Phase III (low-crtcalty or hgh-crtcalty mode): In ths phase, the current ob of τ has ether fnshed or exceeded ts -WCET, so t s known whether the task group s n lowor hgh-crtcalty mode. If the ob has ndeed fnshed, all the low-crtcalty tasks are scheduled; otherwse the group s n hgh-crtcalty mode, and the low-crtcalty tasks contnue to be suspended. When the hgh-crtcalty ob completes (n phase II or III) and the task group has not yet used up ts budget, any suspended low-crtcalty obs wll contnue to be scheduled; however, f they do not complete by the end of the round (.e., when a new hgh-crtcalty ob s released) or by ther deadlnes, whchever s sooner, they wll be dropped. Schedulng mechansm for each round. Recall that the budget of the task group T G n each TG-perod s B and the number of task group budget replenshments n any perod of τ (.e., n each schedulng round of T G ) s h. Wthout loss of generalty, we suppose that Phase I lasts for k TG-perods,.e., the current ob of τ ether swtches to the hgh-crtcalty mode or has completed ts executon n the (k + )th TGperod, where k h. In each schedulng round, tasks are scheduled based on ther pre-computed budget values, x,

4 b and b ( n) as follows. (The method for computng these parameters s presented n Secton IV-D.) Phase I (the frst k TG-perods): To acheve the lowcrtcalty guarantee, n each of the frst k TG-perods, tasks are scheduled as follows: f there are some pendng lowcrtcalty obs, the current ob of τ s executed frst for a specfc perod of tme x (at most) and then the lowcrtcalty obs are executed n ascendng order of the task ndex, where each low-crtcalty task τ s allocated a budget of b. If there are no pendng low-crtcalty obs, the current ob of τ s executed untl t completes or the budget of the task group s used up. The budget parameters should satsfy two condtons: Frst, the total budget allocated to all tasks n the task group should not exceed the budget B of the task group,.e., x + n = b B () Second, snce the current ob of the hgh-crtcalty task s always n the low-crtcalty mode n ths phase, the total budget receved by τ n ths phase should not exceed ts -WCET,.e., k x C τ () () Phase II (the (k + )th TG-perod): To acheve the hghcrtcalty guarantee, the current ob of τ s executed frst wth an allocated budget of x. If ths ob completes ts executon, the server wll schedule the low-crtcalty tasks n the same manner as n the prevous phase. Otherwse, the ob wll swtch from the low-crtcalty mode to the hgh-crtcalty mode and wll contnue to be executed untl t completes or the budget of the task group s used up; n the former case, the server wll schedule the low-crtcalty obs n the same manner as before untl the budget of the task group s used up (some low-crtcalty obs may be dropped). As above, nequalty () should hold to ensure the budget x for τ and the budget b for each low-crtcalty task τ. Moreover, snce the current ob of τ wll swtch to the hghcrtcalty mode f t cannot complete wth the budget x, the budget receved by τ before the mode swtch should be at least equal to ts -WCET. Therefore, (k + ) x C τ () (3) Phase III (the last h k TG-perods): To acheve the hgh-crtcalty guarantee, the schedulng n each of the last h k perods of the task group s as follows: f the current ob of τ has not completed n the consderng TGperod, t wll be executed untl t completes or the budget of the task group s used up. Otherwse, the low-crtcalty tasks are scheduled n ascendng order of the task ndex, where each task τ s allocated a budget of b. Because all budget of the task group wll be allocated to the low-crtcalty tasks n the task group f the task group s n the low-crtcalty ob mode n ths last phase, we can allocate more budget to each low-crtcalty task. In other words, b b (4) Further, the total budget allocated to all the low-crtcalty tasks should not exceed the budget of the task group,.e., n b B () = TABLE I: An example task group: task parameters Task ζ C () C () T U () U () τ τ.8.8 τ Fg. : A schedule of the example n the low-crtcalty mode Fg. 3: A schedule of the example n the hgh-crtcalty mode. We llustrate the schedulng mechansm usng an example. Example. Consder the task group T G depcted n Table I, wth perod T T G = GCD(,3,3) =. Suppose the budget gven to T G s B =.8 and the schedulng parameters are k =, x =, b =, b =., b = and b =. Fgures and 3 each show two schedulng rounds for T G : one n whch the hgh-crtcalty task always exhbts low-crtcalty behavor, and one n whch t exhbts hgh-crtcalty behavor. Note that each round contans h = T τ /T T G = 3 TG-perods and conssts of only Phase II and Phase III (snce k = ). Frst, consder Fgure. In the frst TG-perod (Phase II), the hgh-crtcalty task s frst executed for x = tme unts and completes; hence, the low-crtcalty task τ s then executed for b = tme unts (τ s not executed snce b = ). Snce the task group s n low-crtcalty mode, the low-crtcalty tasks τ and τ are executed n the second and thrd TG-perod (Phase III) for b =. and b = tme unts, respectvely. The second schedulng round works smlarly, except that n the last TG-perod, τ fnshes ts executon before usng all of ts allocated budget. Now consder Fgure 3. Here, τ exceeds ts -WCET n the frst round, so the task group s swtched to hgh-crtcalty mode, and both low-crtcalty tasks are suspended (untl τ completes). In the second round, however, τ remans n lowcrtcalty mode, so τ and τ are executed as usual, based on ther allocated budgets. (Note that n the frst phase of ths round,.e., tme nterval [3,4], nether τ nor τ s executed. Ths s because the current ob of τ has been dropped at tme 3 when τ s released, whereas the budget allocated to τ s b =.) We observe that only some obs of the low-crtcalty tasks mss ther deadlnes: the low-crtcalty obs released n the frst round are suspended when τ s n the hgh-crtcalty mode, but the low-crtcalty obs released n the second round are stll schedulable. We can also valdate that the task group s mxed-crtcalty schedulable n the example schedules.

5 C. Schedulablty Analyss Consder any task group T G = { τ, τ,..., τ n, τ }. We frst establsh the mxed-crtcalty schedulablty condtons for T G under the three-phase schedulng strategy wth a gven budget B and gven schedulng parameters k,x,b,b ( n), where x,b and b are non-negatve real numbers and k s a non-negatve nteger. We then derve the schedulablty condtons for a set of task groups on a processor. Recall that l = T τ /T T G s the number of task group budget replenshments durng each perod of τ, and h = /T T G s the number of task group budget replenshments T τ n each perod of the hgh-crtcalty task τ (.e., n a schedulng round). Lemma IV. gves the schedulng condtons for each low-crtcalty task τ : Lemma IV.. Each low-crtcalty task τ ( n) can be successfully scheduled f all of the nequaltes () () hold, where the nequalty () s gven by N maxol b + (l NmaxOL ) b C τ (), () where N maxol = l h (k + ) + mn{l mod h,k + } s the maxmum number of TG-perods n a perod of τ that overlap wth phases I and II of a schedulng round. Example. Consder the task group depcted n Table II, wth budget B =, k =, b = and b =. The maxmum number of TG-perods that overlap wth the frst k + = TG-perods durng a perod of the hgh-crtcalty task s mn{3 mod, + } =. As demonstrated n Fgure 4, there are two overlappng TG-perods for the st and 4th releases of τ, one for the nd and 3rd releases, and zero for the th release. We can valdate that all obs of τ can fnsh before ther deadlnes. TABLE II: An example task group: task parameters Task ζ C () C () T U () U () τ. τ 3.. N overlap = N overlap N overlap = = N overlap = N overlap = Fg. 4: Number of overlappng TG-perods n Example. Lemma IV.. Every ob of the hgh-crtcalty task τ receves a budget of at least C τ () when t exhbts lowcrtcalty behavor and at least C τ () when t exhbts hgh- crtcalty behavor by ts deadlne f the nequaltes (), (3) and the followng nequalty hold: k x + (h k ) B C τ () (7) The next theorem s derved drectly from Lemmas IV. and IV.. Theorem IV.. Suppose each task group T G = { τ, τ,..., τn, τ } s guaranteed to receve a budget of B n each of ts perod. Then, T G s mxed-crtcalty schedulable under the three-phase schedulng strategy wth schedulng parameters k,x,b and b ( n) f all of the nequaltes () (7) are satsfed. Remarks. We note that Theorem IV. holds regardless of when exactly the task group s scheduled, because all tasks are guaranteed to receve ther allocated budgets as long as the task group as a whole receves a budget of B n each perod. In addton, the hgh-crtcalty behavor of a ob of τ only affects some low-crtcalty obs that need to be executed durng the perod of ths τ ob but not the ones that are released after ts deadlne. Moreover, snce each round of a task group corresponds to a ob of ts hgh-crtcalty task, each task group can swtch from the hgh-crtcalty mode to the low-crtcalty mode ndependently of each other, and the system can swtch to the low-crtcalty mode once all hghcrtcalty obs n the system exhbt low-crtcalty behavor. Let S k be the set of task groups scheduled on a processor k. Recall that each task group T G S k s scheduled wth other task groups n S k as a conventonal mplct-deadlne perodc task wth perod T T G and executon tme B under EDF. Based on the schedulable utlzaton bound of EDF [9], all task groups n S k are schedulable under EDF f ther total utlzaton s less than or equal to,.e., U k = T G S k B /T T G. In other words, every task group T G s guaranteed a budget of B n each of ts perod f U k. The next theorem follows drectly from ths observaton and Theorem IV.. Theorem IV.. Let S k be the set of task groups scheduled on processor π k (accordng to some task groupng algorthm). If for all T G = { τ, τ,..., τ n, τ } n S k, there exst a nonnegatve budget value B R + and non-negatve schedulng parameters k N (k < h ) and x,b,b R + for all n such that U k = T G S k B /T T G and the nequaltes () (7) are satsfed, then all the tasks assgned to processor k are mxed-crtcalty schedulable. We note that, although Theorem IV. only gves a suffcent analyss, our smulaton shows that our parttoned schedulng based on task groupng outperforms an exstng parttoned algorthm based on EDF-VD, and EDF-VD has been shown to be optmal (f all deadlnes are reduced by the same rato) on sngle processors for two-level MC systems from the perspectve of speedup bound []. Next, we present an algorthm based on Mxed Integer Nonlnear Programmng (MINLP) for groupng tasks nto task groups, as well as for computng the budget and schedulng parameters of each task group. D. Mxed Integer Nonlnear Programmng for Task Groupng Let τ(π k ) = { τ,...,τ n,τ },...,τ m be the set of tasks that are assgned to the processor π k based on some tasks-toprocessors packng algorthm. The goal of the task groupng s to determne (a) for each hgh-crtcalty task τ ( m) a subset of the low-crtcalty tasks n τ(π k ) that wll be grouped wth τ nto the task group T G, and (b) for each obtaned task group T G, the values of the task group s budget B and the followng schedulng parameters: k, the number of TGperods n schedulng Phase I; x, the schedulng budget of the hgh-crtcalty task τ ; and b and b, the schedulng budgets for each low-crtcalty task τ n T G. Recall that the perod T T G of T G s a common dvsor of the perods of all tasks

6 n T G ; to reduce the computaton complexty, we fx T T G to be the greatest common dvsor of all tasks n τ(π k ). Thus, l = T τ /T T G and h = T τ /T T G also become constants. To ths end, we defne a MINLP for task groupng based on the schedulablty condtons of the task groups. Snce the task groups T G ( m) are scheduled under EDF as conventonal mplct-deadlne perodc tasks wth perods T T G and budgets B, the smaller ther total utlzaton m = B /T T G s, the better ther schedulablty wll be. Therefore, the obectve of the proposed MINLP s to mnmze the total utlzaton, m = B /T T G, of the task groups so as to mprove ther schedulablty. The constrants of the MINLP are derved drectly from the mxed-crtcalty schedulablty condtons of the task groups (c.f. Theorem IV.). To enable a low-crtcalty task to use the overbookng of multple hgh-crtcalty tasks, we allow each low-crtcalty task τ to appear n multple task groups,.e., each ob of τ can receve budget from more than one task group and the ob s executed whenever t s scheduled n a task group that t belongs. Snce all task groups are executed sequentally on a sngle processor, τ s never executed at the same tme by more than one task group. Hence, f the total budget that τ s guaranteed to receve n each of ts perods from all of ts task groups s greater than or equal to ts -WCET (= -WCET), then τ s schedulable. Based on the above dea, we use m pars of non-negatve real varables (b,,b, ) for each low-crtcalty task τ ( n, m) to represent the budgets that τ receves from task group T G n the three schedulng phases of T G (nstead of usng a sngle par of varables (b,b )). These varables are also used to ndcate whether τ belongs to the task group T G : f b, = b, =, then τ does not belong to T G, otherwse t does. Based on Theorem IV., the task set τ(π k ) s mxedcrtcalty schedulable f all of the nequaltes () (7) are satsfed for every task group T G. Due to the above encodng of low-crtcalty tasks, the constrants for each task group T G can be obtaned by replacng b wth b,, b wth b,, N maxol wth N, maxol n the nequaltes and rewrttng the nequalty (7) as m ( = b, N, maxol + b, (l N, maxol ) ) C τ (). To reduce the computaton complexty, nstead of usng the nequalty () (.e., x + n = b, ), we set B = x + n = b,. By mergng all constrants, we obtan the MINLP formulaton shown n Fgure. Constrants () (7) correspond to nequaltes () (7), and the remanng constrants are condtons on the varables, as was dscussed earler. Gven an MC task set, f the total task group utlzaton obtaned from soluton of the MINLP s more than one, then ths task set s unschedulable by the task group based schedulng and wll be reected by the scheduler. Complexty. If n and m are the number of low- and hghcrtcalty tasks n the task set, respectvely, then the MINLP has mn + m constrants, m nteger varables and m + mn real varables. Hence, the number of varables and constrants s polynomally bounded n the sze of the nput problem (.e., the task set sze), and t can be solved by a fully polynomaltme approxmaton scheme. The latter s therefore suffcent to obtan all the schedulng parameters. Mnmze m = (( n = b, + x )/T T G ) Subected to () B = x + n = b, () k x C τ () (3) (k + ) x C τ () ( =,...,m) ( =,...,m) ( =,...,m) (4) b, b, ( =,...,n; =,...,m) () n = b, B ( =,...,m) () m (, = b N, maxol + b, (l NmaxOL, ) ) C τ () ( =,...,n; =,...,m) (7) k x + (h k ) B C τ () ( =,...,m) (8) k h ( =,...,m) (9) x R + ( =,...,m) () b, R + ( =,...,n; =,...,m) () b, R + ( =,...,n; =,...,m) () k N ( =,...,m) Fg. : MINLP formulaton of the task groupng Dscussons. The MINLP formulaton gven n Fgure can easly be modfed to consder other types of constrants as well e.g., constrants on the number of task groups that each low-crtcalty task can belong, constrants on the number of low-crtcalty tasks n each task group, etc. Due to space constrants, we omt them here. E. EDF-based Two-phased Schedulng Strategy The proposed budget-drven three-phased schedulng algorthm wthn a task group (c.f. Secton IV-B) may lead to hgh context swtch overhead when there s a large number of low-crtcalty tasks n a task group. To avod ths drawback, we propose a two-phased varant of the algorthm that schedules lowcrtcalty tasks based on EDF. (Ths verson of the algorthm s used n our evaluaton.) Gven a task group T G wth budget B and schedulng parameters k and x, the schedulng durng each round (.e., perod of the hgh-crtcalty task τ n the task group) works as follows: Phase I (the frst k ( k h ) TG-perods): In each TG-perod, f there are some pendng low-crtcalty obs n the task group, the current ob of τ s executed frst for a specfc perod of up to x, and then the low-crtcalty obs n the task group are executed under the EDF polcy. Otherwse, the current ob of τ s executed untl t completes or the budget of the task group expres. Phase II (the last h k perod(s) of task groups): In each TG-perod, the ob of τ s executed frst untl t fnshes and then the pendng low-crtcalty obs n the task group are executed under the EDF polcy. If the current ob of τ exhbts hgh-crtcalty behavor, the low-crtcalty obs that cannot fnsh when ther deadlnes are reached or when a new ob of τ s released are dropped. Note that, unlke n the budget-drven three-phased schedulng strategy, the (k + )th TG-perod and the last h k TGperods are consdered as one phase n the EDF-based twophased schedulng strategy. The reason for ths s that under the three-phased strategy, dfferent budgets (b and b ) may be used for each low-crtcalty task τ n the (k + )th TGperod and n the last h k TG-perods, whereas under the EDF-based schedulng strategy, low-crtcalty tasks are always scheduled usng the EDF polcy. The schedulablty under the EDF-based two-phased schedulng strategy s gven by the next theorem.

7 Theorem IV.3. If a task group T G = { τ, τ,..., τ n, τ } s mxed-crtcalty schedulable under the three-phased schedulng strategy wth budget B and schedulng parameters k,x,b,b, then T G s also mxed-crtcalty schedulable under the EDF-based two-phased schedulng strategy wth budget B and schedulng parameters k and x. Example 3. Consder the task group shown n Table III, wth perod, budget.9, k = and x =. As llustrated n Fgure, all low-crtcalty tasks can be successfully scheduled by the three-phased server wth budgets b =, b =., b =. and b =. From Fgure 7, we can see that an EDF-based schedule for the low-crtcalty tasks can be obtaned by nterchangng some executons of τ and τ, resultng n a smaller number of context swtches. TABLE III: An example task group: task parameters Task ζ C () C () T U () U () τ τ.8.8 τ Fg. : A budget-drven schedule Fg. 7: An EDF-based schedule obtaned by transposton. Due to Theorem IV.3, we can determne the task groups and ther schedulng parameters for the EDF-based two-phased schedulng algorthm by solvng the MINLP formulated n the prevous secton. A smple case study that llustrates the endto-end behavor and the benefts of our task-groupng-based schedulng algorthm compared to an exstng algorthm based on vrtual deadlne [7] s provded n [8]. Remarks. The context swtch overhead of EDF-based twophased schedulng can further be mtgated by consderng the slcng number of low-crtcalty tasks as a constrant n the MINLP formulaton and by extendng analyss to allow other values for the task group perod and/or to determne the tme of the task group budget replenshment dynamcally. In addton, we can also enable servce guarantee for low-crtcalty tasks usng a constrant on the number of task groups that each lowcrtcalty task can belong n the MINLP formulaton. V. TASK-GROUP-BASED PARTITIONED MC SCHEDULING In ths secton, we present a parttonng algorthm for assgnng a set of mxed-crtcalty tasks τ to processors. We consder a platform wth p dentcal, unt-capacty processors, denoted as π = {π,..., π p }. Our algorthm ams to balance the loads across all processors, whle selectng hgh-crtcalty (lowcrtcalty) tasks to assgn based on decreasng hgh-crtcalty (low-crtcalty) utlzaton. Let τ = {τ,...,τ n } be the set of hgh-crtcalty tasks and τ = {τ n +,...,τ n } be the set of low-crtcalty tasks of the task set τ. Algorthm s our parttonng algorthm, wrtten n pseudocode. In ths algorthm, the set of tasks allocated to each processor π s denoted by τ(π ). The low-crtcalty utlzaton U (τ ) and the hgh-crtcalty utlzaton U (τ ) of a task τ are defned as the task s -WCET and -WCET dvded by ts perod, respectvely. For brevty, we smply refer to the hgh-crtcalty utlzaton of a hgh-crtcalty task (and, smlarly, the low-crtcalty utlzaton of a low-crtcalty task) as that task s utlzaton. Further, the cumulatve task group utlzaton for a set of tasks s the total utlzaton (.e., budget dvded by perod) of all task groups of the task set (determned usng the MINLP method n Secton IV-D). Algorthm Task group based MC parttoned algorthm. : τ(π ) /, for all =,..., p. : Sort hgh-crtcalty tasks τ n decreasng order of hgh-crtcalty utlzaton and then n decreasng order of low-crtcalty utlzaton. 3: Sort low-crtcalty tasks τ n decreasng order of low-crtcalty utlzaton and then n ncreasng order of perod. 4: whle τ / and τ / do : τ selected NIL : f τ / then 7: τ selected The frst task n τ 8: f τ / then 9: τ selected The frst task n τ : f τ selected ( = NIL or τselected NIL and U (τ selected ) < U (τ selected )) then : τ selected τ selected : Remove τ selected from ts set 3: Fnd a processor π π that has a mnmal cumulatve task group utlzaton for the tasks τ(π ) {τ selected } 4: f the cumulatve task group utlzaton for tasks τ(π ) {τ selected } does not exceed then : τ(π ) τ(π ) {τ selected } : else 7: return FAILURE 8: return SUCCESS The algorthm starts by ntalzng τ(π ) to null (Lne ) and by sortng hgh-crtcalty tasks and low-crtcalty tasks (Lnes 3). Hgh-crtcalty tasks are sorted n decreasng order of hgh-crtcalty utlzaton, and tasks wth the same hghcrtcalty utlzaton are further ordered by decreasng lowcrtcalty utlzaton (Lne ). Low-crtcalty tasks are sorted n decreasng order of low-crtcalty utlzaton, and tasks wth the same low-crtcalty utlzaton are further ordered by ncreasng perod (Lne 3). Our reason for ths sortng s as follows: for hgh-crtcalty tasks wth the same hgh-crtcalty utlzaton, a task wth a hgher low-crtcalty utlzaton typcally has less overbookng; and for low-crtcalty tasks wth the same low-crtcalty utlzaton, the task wth a smaller perod tends to have a lower utlzaton rate of overbookng, snce the task group perod s chosen as the common dvsor of the perods of the tasks wthn the task group. However, the algorthm can easly be extended to other sortng methods (whch we plan to nvestgate n future work). The algorthm then selects the task τ selected that has the largest utlzaton (Lnes ), and removes t from ts task set (Lne ). Next, the algorthm selects a processor π for ths task, such that the cumulatve task group utlzaton for ths task along wth the exstng tasks on ths processor s mnmal (Lne 3); ths s done to balance the load across

8 dfferent processors. If the resultng mnmum cumulatve task group utlzaton on the selected processor π does not exceed one, τ selected s assgned to π (Lnes 4 ). Otherwse, the algorthm aborts wth a falure (Lne 7). If every task s successfully allocated to a processor, the algorthm reports success (Lne 8). Based on Theorem IV., the resultng task allocaton obtaned by Algorthm always ensures that tasks on each processor can be successfully scheduled by the task group based schedulng strategy. It should be noted that the crtcalty levels of the tasks on a processor may be the same; n ths case, all tasks on the processor are drectly scheduled under EDF. VI. EVALUATION In ths secton, we expermentally compare the performance of our proposed scheme, TG-PEDF, wth the followng parttoned mxed-crtcalty schedulng algorthms: DC-RM: a parttoned schedulng algorthm from [9] that s based on RM prorty assgnment; DC-Audsley: a parttoned schedulng algorthm from [9] that s based on Audsley s optmal prorty assgnment; MC-Partton: a parttoned schedulng algorthm based on EDF-VD (from []); EY-FF: a straghtforward extenson of EY [] for parttoned schedulng wth the FF packng strategy from [4]; MPVD: an extenson of EY for parttoned schedulng wth the hybrd packng strategy from [4]; MPVD-HA: MPVD wth the heavy low-crtcalty task aware allocaton polcy from [4]; and MPVD-HA-BF: MPVD-HA wth the optmzed vrtual deadlne tunng from [4]. Obectves. We have three man goals for our evaluaton: () to compare the performance of TG-PEDF n terms of mxedcrtcalty schedulablty to that of the above algorthms; () to evaluate how well TG-PEDF can protect low-crtcalty obs when a hgh-crtcalty task s n hgh-crtcalty mode; (3) to evaluate the deadlne mss ratos of low-crtcalty tasks under dfferent frequences of hgh-crtcalty tasks beng n hghcrtcalty mode; (4) to evaluate the computatonal complexty of TG-PEDF; and () to evaluate the effect of overbookng by allowng a low-crtcalty task to appear n multple task groups n terms of schedulablty on a unprocessor. For Obectve (), our metrc s the mpact rato of lowcrtcalty tasks wth respect to dfferent percentages of hghcrtcalty tasks beng n the hgh-crtcalty mode. We defne the mpact rato of low-crtcalty tasks as the fracton of such tasks for whch at least one ob s dropped or msses ts deadlne because a hgh-crtcalty task s n hgh-crtcalty mode. The deadlne mss rato of low-crtcalty obs s defned as the fracton of low-crtcalty obs that mss ther deadlnes. Due to space constrants, some addtonal results are presented n our techncal report [8]. A. Workload For our experments, we used randomly generated task sets, whch we created usng the algorthm from [7]. We set the probablty pcrtcalty that a task s a hgh-crtcalty task to.. The perod (relatve deadlne) T of each task τ was an nteger that was drawn unformly at random from the nterval [, ]. The low-crtcalty WCET C () of task τ was drawn from [. T, T ] and, f τ was a hghcrtcalty task, ts hgh-crtcalty WCET C () was drawn from [ C (), 4 C ()]. We defne the normalzed average utlzaton U avg (τ) of a task set τ to be: U avg (τ) = U (τ) +U (τ) (8) p where U (τ) and U (τ) are the cumulatve low-crtcalty utlzaton and the cumulatve hgh-crtcalty utlzaton of a task set τ, respectvely. p s the number of processors. Usng the above parameters, we generated the tasks one at a tme untl the followng condtons on system utlzaton were satsfed: () Uavg. U avg (τ) Uavg +.; () U (τ) p; and () U (τ) p, where Uavg {,,.7,.7,.8,.8,.9,.9,.97} and p {4,8,}. For each Uavg and each p value, we generated, task sets for each of the above eght algorthms. Snce the evaluaton of the mpact rato of low-crtcalty tasks and the deadlne mss rato of low-crtcalty obs nvolves expensve smulatons, we used for ths purpose task sets randomly chosen from the generated task sets wth Uavg {,.7,.8,.9} that can be scheduled by TG-PEDF. We ran each smulaton for,, tme unts. To evaluate the effect of resource overbookng on the schedulablty for dfferent maxmum numbers of hgh-crtcalty task groups that a low-crtcalty task can belong to, we set pcrtcalty to.8, and we randomly drew the low-crtcalty task utlzaton from [., ] to generate more hgh-crtcalty tasks. B. Results Mxed-crtcalty schedulablty. Fgures 8a 8c show the fracton of schedulable task sets versus the normalzed average utlzaton of the dfferent algorthms n 4-processor, 8-processor, and -processor systems, respectvely. We begn wth a few observatons about the exstng algorthms. Frst, the results for 8- and -processor systems show that the performance of EY-FF decreased as the number of processors ncreased. Ths s because the space for EY to tune the vrtual deadlnes s reduced due to the unbalanced allocaton of hgh-crtcalty tasks. Second, the results n Fgure 8c show that, n -processor systems, MPVD, MPVD-HA and MPVD-HA-BF outperformed EY-FF by balancng the demand of hgh-crtcalty tasks among dfferent processors. MPVD- HA enhances MPVD by assgnng the heavy low-crtcalty tasks to processors before the hgh-crtcalty ones, but the performance of the two algorthms s smlar when the task sets contan few low-crtcalty tasks. MPVD-HA-BF ams to mprove schedulablty usng a new vrtual deadlne tunng algorthm, and ts schedulablty s ndeed better than MPVD and MPVD-HA n most cases; however, the vrtual deadlne tunng algorthm does not take the low-crtcalty tasks nto account, so t can affect schedulablty even f the demand n low-crtcalty mode s low. For ths reason, MPVD-HA-BF dd not outperform EY-FF on the systems wth 4 and 8 processors, and t performed worse than MPVD and MPVD-HA on - processor systems wth hgher average utlzaton. Our man result s that TG-PEDF consstently outperformed all exstng parttoned MC schedulng algorthms. The performance gap tends to wden as the average utlzaton of the

9 Fracton of schedulable task sets DC RM DC AUDSLEY MC PARTITION EY FF MPVD MPVD HA MPVD HA BF TG PEDF Fracton of schedulable task sets DC RM DC AUDSLEY MC PARTITION EY FF MPVD MPVD HA MPVD HA BF TG PEDF Fracton of schedulable task sets DC RM DC AUDSLEY MC PARTITION EY FF MPVD MPVD HA MPVD HA BF TG PEDF Normalzed average utlzaton (a) 4-processor systems Normalzed average utlzaton (b) 8-processor systems Fg. 8: Fracton of mxed-crtcalty schedulable task sets for dfferent algorthms Normalzed average utlzaton (c) -processor systems Impact rato of low crtcalty tasks U avg = U avg =.7 U avg =.8. U avg = Rato of hgh crtcalty tasks wth hgh crtcalty behavor Fg. : Impact rato of low-crtcalty tasks. Deadlne mss rato of low crtcalty obs U = avg U =.7 avg U =.8 avg U avg = Hgh crtcalty ob behavor probablty Fg. : Deadlne mss rato of low-crtcalty obs. Fracton of schedulable task sets N max sub = N max sub = N sub = 3. N sub = Normalzed average utlzaton Fg. 3: Effect of overbookng on schedulablty. task sets ncreases; the reason s that, as the number of tasks ncreases, there are more and more opportuntes for TG-PEDF to fnd an approprate task group for each low-crtcalty task. Impact of hgh-crtcalty behavor on low-crtcalty tasks. Fgure shows the mpact rato of the low-crtcalty tasks for dfferent percentages of hgh-crtcalty tasks wth hghcrtcalty behavor on 4-processor systems. The results show that only a certan fracton of low-crtcalty tasks are affected, and that ths fracton grows wth the fracton of hghcrtcalty tasks that exhbt hgh-crtcalty behavor. Ths s because, wth TG-PEDF, the hgh-crtcalty behavor of a hgh-crtcalty task can only affect low-crtcalty tasks that are n the same task group wth ths hgh-crtcalty task, but not low-crtcalty tasks n other task groups. We make three addtonal observatons. Frst, the relatonshp s not exactly lnear because some low-crtcalty tasks are grouped together wth more than one hgh-crtcalty task, and can thus be affected by any one of these tasks. Second, when all hgh-crtcalty tasks exhbt hgh-crtcalty behavor (.e., the rato of hgh-crtcalty tasks wth hghcrtcalty behavor s one), there are stll some low-crtcalty tasks that are not affected. The reason for ths s that there can be processors wth only low-crtcalty tasks and no hgh-crtcalty tasks, and these cannot be nfluenced by the hgh-crtcalty behavor on other processors. Fnally, the results are not senstve to dfferent U avg values (the curves correspondng to dfferent system utlzaton values cross). Ths s expected because the fracton of low-crtcalty tasks that are affected depends prmarly on the task groupng and the fracton of hgh-crtcalty tasks that exhbt hgh-crtcalty behavor nstead of the system utlzaton. Real-tme performance of low-crtcalty tasks. Fgure shows the effect of hgh-crtcalty ob behavor on the deadlne mss rato of low-crtcalty obs, usng the 4-processor system as an example. Not surprsngly, there are no deadlne msses when the probablty of hgh-crtcalty ob behavor s zero: when the system exhbts low-crtcalty behavor, the taskgroup-based schedulng can ensure that the system s schedulable. As the probablty of hgh-crtcalty ob behavor rses, the deadlne mss rato rses as well, but only gradually; the reason s, agan, that the hgh-crtcalty behavor of a ob can only affect low-crtcalty obs wth whch t shares a task group, but not low-crtcalty obs n other task groups. Addtonally, snce the solaton between task groups enables the system to swtch back to low-crtcalty mode once all hgh-crtcalty tasks exhbt low-crtcalty ob behavor (or, at a task group level, even when only the local hgh-crtcalty task exhbts low-crtcalty behavor), the deadlne mss rato of the lowcrtcalty obs can be kept low even for long-runnng systems. We also observe that the deadlne mss rato curves correspondng to dfferent averge system utlzaton values cross each other,.e., a task set wth a hgher average utlzaton may have a smaller deadlne mss rato of low-crtcalty obs than that of a task set wth a lower average utlzaton. Ths s not surprsng, because the deadlne mss rato of lowcrtcalty obs can become smaller as the total number of low-crtcalty obs ncreases, whch can be the case when the average utlzaton ncreases. It s worth notng that the exstng mxed-crtcalty algorthms drop all of the low-crtcalty tasks n the system as soon as a hgh-crtcalty ob exhbts hgh-crtcalty behavor, so they provde no real-tme performance guarantee for low-crtcalty tasks n the hgh-crtcalty mode. Effect of resource overbookng on schedulablty. Fgure 3 llustrates the effect of enablng low-crtcalty tasks to use the overbookng of multple hgh-crtcalty tasks on the schedu-

10 lablty of the task set on a unprocessor. Here, Nsub max s the maxmum number of task groups for each low-crtcalty task. The results show that the more hgh-crtcalty tasks that a lowcrtcalty task s executed wth, the better the schedulablty. In addton, the schedulablty gap tends to wden as the average utlzaton of the task set ncreases, whch further ndcates the effectveness of usng resource overbookng of hgh-crtcalty tasks. We note that ths can also lead to more hgh-crtcalty tasks nterferng wth a low-crtcalty task; however, snce the low-crtcalty schedulablty guarantee also reflects the schedulablty of low-crtcalty tasks, the resultng benefts outwegh the addtonal nterference. Computaton complexty. We also evaluated the tme needed per task set to group the tasks and to compute the task groups schedulng parameters n our experments, when usng the APMontor optmzaton sute [3] for solvng the MINLP problem. The results show that the formulated MINLP problem can be solved effcently (by a fully polynomal-tme approxmaton scheme): t took less than ms across all task sets n our experments. VII. CONCLUSION We have proposed TG-PEDF, a parttoned schedulng technque for mxed-crtcalty multprocessor real-tme systems that s based on task groupng. TG-PEDF works by assocatng each hgh-crtcalty task wth a subset of the low-crtcalty tasks and by encapsulatng them n separate task groups. Task groups are scheduled wth other task groups usng the EDF polcy, whle the tasks wthn each group are scheduled usng a server-based strategy that can ensure the mxed-crtcalty guarantees. Ths strategy not only guarantees solaton among hgh-crtcalty tasks, t also enables more low-crtcalty tasks to meet ther deadlnes. TG-PEDF s also compostonal: new tasks can easly be composed wth exstng ones va task groups. We have presented a schedulablty analyss for the system under the proposed schedulng strategy and an MINLP formulaton for the task groupng, as well as a packng algorthm that consders crtcalty n task deployment to optmze resource use. Our evaluaton shows that TG-PEDF consstently outperforms exstng multprocessor MC schedulng algorthms n terms of system schedulablty, whle enablng better tmng performance for low-crtcalty tasks. As a future drecton, we plan to mplement the proposed technque on a real platform and study ts real-tme performance and schedulng overhead, as well as to extend t to multple crtcalty levels. ACKNOWLEDGMENTS The authors would lke to thank Arvnd Easwaran, Nan Guan and Chuanca Gu for ther nputs on the mplementaton of exstng algorthms, Andreas Haeberlen for hs useful feedbacks on the paper, and Guowe Wu for hs support. Ths work was supported n part by the NSF grants CNS 78, ECCS 33 and CNS 39984, the ONR grant N , the Natonal Natural Scence Foundaton of Chna under Grant No and 44, and the Fundamental Research Funds for the Central Unverstes (No. DUT3JS). REFERENCES [] S. Vestal, Preemptve schedulng of mult-crtcalty systems wth varyng degrees of executon tme assurance, n RTSS, 7. [] D. de Nz, K. Lakshmanan, and R. Rakumar, On the schedulng of mxed-crtcalty real-tme task sets, n RTSS, 9. [3] H. L and S. Baruah, Load-based schedulablty analyss of certfable mxed-crtcalty systems, n Proceedngs of the tenth ACM nternatonal conference on Embedded software,. [4] S. K. Baruah, V. Bonfac, G. DAngelo, A. Marchett-Spaccamela, S. Van Der Ster, and L. Stouge, Mxed-crtcalty schedulng of sporadc task systems, n Algorthms ESA,. [] N. Guan, P. Ekberg, M. Stgge, and W. Y, Effectve and effcent schedulng of certfable mxed-crtcalty sporadc task systems, n RTSS,. [] S. Baruah, V. Bonfac, G. D Angelo, H. L, A. Marchett-Spaccamela, S. Van Der Ster, and L. Stouge, The preemptve unprocessor schedulng of mxed-crtcalty mplct-deadlne sporadc task systems, n ECRTS,. [7] A. Easwaran, Demand-based schedulng of mxed-crtcalty sporadc tasks on one processor, n RTSS, 3. [8] M. S. Mollson, J. P. Erckson, J. H. Anderson, S. K. Baruah, and J. A. Scoredos, Mxed-crtcalty real-tme schedulng for multcore systems, n CIT,. [9] O. R. Kelly, H. Aydn, and B. Zhao, On parttoned schedulng of fxed-prorty mxed-crtcalty task sets, n TrustCom,. [] S. Baruah, B. Chattopadhyay, H. L, and I. Shn, Mxed-crtcalty schedulng on multprocessors, Real-Tme Systems, 4. [] D. de Nz and L. T. X. Phan, Parttoned schedulng of mult-modal mxed-crtcalty real-tme systems on multprocessor platforms, n RTAS, 4. [] ISO/DIS road vehcles - functonal safety, [3] A. Burns and R. Davs, Mxed crtcalty systems: A revew, Department of Computer Scence, Unversty of York, Tech. Rep, 3. [4] S. K. Baruah, A. Burns, and R. I. Davs, Response-tme analyss for mxed crtcalty systems, n RTSS,. [] F. Santy, L. George, P. Therry, and J. Goossens, Relaxng mxedcrtcalty schedulng strctness for task sets scheduled wth fp, n ECRTS,. [] H.-M. Huang, C. Gll, and C. Lu, Implementaton and evaluaton of mxed-crtcalty schedulng approaches for perodc tasks, n RTAS,. [7] H. Su and D. Zhu, An elastc mxed-crtcalty task model and ts schedulng algorthm, n DATE, 3. [8] M. Jan, L. Zaourar, and M. Ptel, Maxmzng the executon rate of low crtcalty tasks n mxed crtcalty system, n WMC, 3. [9] A. Burns and S. Baruah, Towards a more practcal model for mxed crtcalty systems, n WMC, RTSS, 3. [] P. Huang, G. Gannopoulou, N. Stomenov, and L. Thele, Servce adaptons for mxed-crtcalty systems, Techncal Report 3, ETH Zurch, Tech. Rep., 3. [] X. Gu, A. Easwaran, K. M. Phan, and I. Shn, Compostonal mxedcrtcalty schedulng, n CRTS, 4. [] R. M. Pathan, Schedulablty analyss of mxed-crtcalty systems on multprocessors, n ECRTS,. [3] S. K. Baruah, Optmal utlzaton bounds for the fxed-prorty schedulng of perodc task systems on dentcal multprocessors, IEEE Transactons on Computers, 4. [4] C. Gu, N. Guan, Q. Deng, and W. Y, Parttoned mxed-crtcalty schedulng on multprocessor platforms, n DATE. IEEE, 4. [] P. Ekberg and W. Y, Boundng and shapng the demand of generalzed mxed-crtcalty sporadc task systems, Real-tme systems, 4. [] H. Su, D. Zhu, and D. Mossé, Schedulng algorthms for elastc mxedcrtcalty tasks n multcore systems, 3. [7] P. Huang, G. Gannopoulou, N. Stomenov, and L. Thele, Servce adaptons for mxed-crtcalty systems, n ASP-DAC, 4. [8] J. Ren and L. T. X. Phan, Mxed-crtcalty schedulng on multprocessors usng task groupng, Techncal Report, Unversty of Pennsylvana, Tech. Rep.,. [Onlne]. Avalable: http: // lnhphan/papers/phan-ecrts-tr.pdf [9] C. L. Lu and J. W. Layland, Schedulng algorthms for multprogrammng n a hard-real-tme envronment, Journal of the ACM, 973. [3] J. D. Hedengren. (4) Apmontoroptmzaton sute. [Onlne]. Avalable:

Maintaining temporal validity of real-time data on non-continuously executing resources

Maintaining temporal validity of real-time data on non-continuously executing resources Mantanng temporal valdty of real-tme data on non-contnuously executng resources Tan Ba, Hong Lu and Juan Yang Hunan Insttute of Scence and Technology, College of Computer Scence, 44, Yueyang, Chna Wuhan

More information

AADL : about scheduling analysis

AADL : about scheduling analysis AADL : about schedulng analyss Schedulng analyss, what s t? Embedded real-tme crtcal systems have temporal constrants to meet (e.g. deadlne). Many systems are bult wth operatng systems provdng multtaskng

More information

Compiler Design. Spring Register Allocation. Sample Exercises and Solutions. Prof. Pedro C. Diniz

Compiler Design. Spring Register Allocation. Sample Exercises and Solutions. Prof. Pedro C. Diniz Compler Desgn Sprng 2014 Regster Allocaton Sample Exercses and Solutons Prof. Pedro C. Dnz USC / Informaton Scences Insttute 4676 Admralty Way, Sute 1001 Marna del Rey, Calforna 90292 pedro@s.edu Regster

More information

Parallelism for Nested Loops with Non-uniform and Flow Dependences

Parallelism for Nested Loops with Non-uniform and Flow Dependences Parallelsm for Nested Loops wth Non-unform and Flow Dependences Sam-Jn Jeong Dept. of Informaton & Communcaton Engneerng, Cheonan Unversty, 5, Anseo-dong, Cheonan, Chungnam, 330-80, Korea. seong@cheonan.ac.kr

More information

6.854 Advanced Algorithms Petar Maymounkov Problem Set 11 (November 23, 2005) With: Benjamin Rossman, Oren Weimann, and Pouya Kheradpour

6.854 Advanced Algorithms Petar Maymounkov Problem Set 11 (November 23, 2005) With: Benjamin Rossman, Oren Weimann, and Pouya Kheradpour 6.854 Advanced Algorthms Petar Maymounkov Problem Set 11 (November 23, 2005) Wth: Benjamn Rossman, Oren Wemann, and Pouya Kheradpour Problem 1. We reduce vertex cover to MAX-SAT wth weghts, such that the

More information

An Optimal Algorithm for Prufer Codes *

An Optimal Algorithm for Prufer Codes * J. Software Engneerng & Applcatons, 2009, 2: 111-115 do:10.4236/jsea.2009.22016 Publshed Onlne July 2009 (www.scrp.org/journal/jsea) An Optmal Algorthm for Prufer Codes * Xaodong Wang 1, 2, Le Wang 3,

More information

Virtual Machine Migration based on Trust Measurement of Computer Node

Virtual Machine Migration based on Trust Measurement of Computer Node Appled Mechancs and Materals Onlne: 2014-04-04 ISSN: 1662-7482, Vols. 536-537, pp 678-682 do:10.4028/www.scentfc.net/amm.536-537.678 2014 Trans Tech Publcatons, Swtzerland Vrtual Machne Mgraton based on

More information

A mathematical programming approach to the analysis, design and scheduling of offshore oilfields

A mathematical programming approach to the analysis, design and scheduling of offshore oilfields 17 th European Symposum on Computer Aded Process Engneerng ESCAPE17 V. Plesu and P.S. Agach (Edtors) 2007 Elsever B.V. All rghts reserved. 1 A mathematcal programmng approach to the analyss, desgn and

More information

Support Vector Machines

Support Vector Machines /9/207 MIST.6060 Busness Intellgence and Data Mnng What are Support Vector Machnes? Support Vector Machnes Support Vector Machnes (SVMs) are supervsed learnng technques that analyze data and recognze patterns.

More information

The Greedy Method. Outline and Reading. Change Money Problem. Greedy Algorithms. Applications of the Greedy Strategy. The Greedy Method Technique

The Greedy Method. Outline and Reading. Change Money Problem. Greedy Algorithms. Applications of the Greedy Strategy. The Greedy Method Technique //00 :0 AM Outlne and Readng The Greedy Method The Greedy Method Technque (secton.) Fractonal Knapsack Problem (secton..) Task Schedulng (secton..) Mnmum Spannng Trees (secton.) Change Money Problem Greedy

More information

An Investigation into Server Parameter Selection for Hierarchical Fixed Priority Pre-emptive Systems

An Investigation into Server Parameter Selection for Hierarchical Fixed Priority Pre-emptive Systems An Investgaton nto Server Parameter Selecton for Herarchcal Fxed Prorty Pre-emptve Systems R.I. Davs and A. Burns Real-Tme Systems Research Group, Department of omputer Scence, Unversty of York, YO10 5DD,

More information

Parallel matrix-vector multiplication

Parallel matrix-vector multiplication Appendx A Parallel matrx-vector multplcaton The reduced transton matrx of the three-dmensonal cage model for gel electrophoress, descrbed n secton 3.2, becomes excessvely large for polymer lengths more

More information

Problem Definitions and Evaluation Criteria for Computational Expensive Optimization

Problem Definitions and Evaluation Criteria for Computational Expensive Optimization Problem efntons and Evaluaton Crtera for Computatonal Expensve Optmzaton B. Lu 1, Q. Chen and Q. Zhang 3, J. J. Lang 4, P. N. Suganthan, B. Y. Qu 6 1 epartment of Computng, Glyndwr Unversty, UK Faclty

More information

Verification by testing

Verification by testing Real-Tme Systems Specfcaton Implementaton System models Executon-tme analyss Verfcaton Verfcaton by testng Dad? How do they know how much weght a brdge can handle? They drve bgger and bgger trucks over

More information

GSLM Operations Research II Fall 13/14

GSLM Operations Research II Fall 13/14 GSLM 58 Operatons Research II Fall /4 6. Separable Programmng Consder a general NLP mn f(x) s.t. g j (x) b j j =. m. Defnton 6.. The NLP s a separable program f ts objectve functon and all constrants are

More information

The Codesign Challenge

The Codesign Challenge ECE 4530 Codesgn Challenge Fall 2007 Hardware/Software Codesgn The Codesgn Challenge Objectves In the codesgn challenge, your task s to accelerate a gven software reference mplementaton as fast as possble.

More information

Adaptive Resource Allocation Control with On-Line Search for Fair QoS Level

Adaptive Resource Allocation Control with On-Line Search for Fair QoS Level Adaptve Resource Allocaton Control wth On-Lne Search for Far QoS Level Fumko Harada, Toshmtsu Usho, Graduate School of Engneerng Scence Osaka Unversty {harada@hopf, usho@}sysesosaka-uacjp Yukkazu akamoto

More information

CHAPTER 2 PROPOSED IMPROVED PARTICLE SWARM OPTIMIZATION

CHAPTER 2 PROPOSED IMPROVED PARTICLE SWARM OPTIMIZATION 24 CHAPTER 2 PROPOSED IMPROVED PARTICLE SWARM OPTIMIZATION The present chapter proposes an IPSO approach for multprocessor task schedulng problem wth two classfcatons, namely, statc ndependent tasks and

More information

Assignment # 2. Farrukh Jabeen Algorithms 510 Assignment #2 Due Date: June 15, 2009.

Assignment # 2. Farrukh Jabeen Algorithms 510 Assignment #2 Due Date: June 15, 2009. Farrukh Jabeen Algorthms 51 Assgnment #2 Due Date: June 15, 29. Assgnment # 2 Chapter 3 Dscrete Fourer Transforms Implement the FFT for the DFT. Descrbed n sectons 3.1 and 3.2. Delverables: 1. Concse descrpton

More information

TPL-Aware Displacement-driven Detailed Placement Refinement with Coloring Constraints

TPL-Aware Displacement-driven Detailed Placement Refinement with Coloring Constraints TPL-ware Dsplacement-drven Detaled Placement Refnement wth Colorng Constrants Tao Ln Iowa State Unversty tln@astate.edu Chrs Chu Iowa State Unversty cnchu@astate.edu BSTRCT To mnmze the effect of process

More information

Real-time Scheduling

Real-time Scheduling Real-tme Schedulng COE718: Embedded System Desgn http://www.ee.ryerson.ca/~courses/coe718/ Dr. Gul N. Khan http://www.ee.ryerson.ca/~gnkhan Electrcal and Computer Engneerng Ryerson Unversty Overvew RTX

More information

An Iterative Solution Approach to Process Plant Layout using Mixed Integer Optimisation

An Iterative Solution Approach to Process Plant Layout using Mixed Integer Optimisation 17 th European Symposum on Computer Aded Process Engneerng ESCAPE17 V. Plesu and P.S. Agach (Edtors) 2007 Elsever B.V. All rghts reserved. 1 An Iteratve Soluton Approach to Process Plant Layout usng Mxed

More information

Dynamic Voltage Scaling of Supply and Body Bias Exploiting Software Runtime Distribution

Dynamic Voltage Scaling of Supply and Body Bias Exploiting Software Runtime Distribution Dynamc Voltage Scalng of Supply and Body Bas Explotng Software Runtme Dstrbuton Sungpack Hong EE Department Stanford Unversty Sungjoo Yoo, Byeong Bn, Kyu-Myung Cho, Soo-Kwan Eo Samsung Electroncs Taehwan

More information

R s s f. m y s. SPH3UW Unit 7.3 Spherical Concave Mirrors Page 1 of 12. Notes

R s s f. m y s. SPH3UW Unit 7.3 Spherical Concave Mirrors Page 1 of 12. Notes SPH3UW Unt 7.3 Sphercal Concave Mrrors Page 1 of 1 Notes Physcs Tool box Concave Mrror If the reflectng surface takes place on the nner surface of the sphercal shape so that the centre of the mrror bulges

More information

Lecture 7 Real Time Task Scheduling. Forrest Brewer

Lecture 7 Real Time Task Scheduling. Forrest Brewer Lecture 7 Real Tme Task Schedulng Forrest Brewer Real Tme ANSI defnes real tme as A Real tme process s a process whch delvers the results of processng n a gven tme span A data may requre processng at a

More information

Subspace clustering. Clustering. Fundamental to all clustering techniques is the choice of distance measure between data points;

Subspace clustering. Clustering. Fundamental to all clustering techniques is the choice of distance measure between data points; Subspace clusterng Clusterng Fundamental to all clusterng technques s the choce of dstance measure between data ponts; D q ( ) ( ) 2 x x = x x, j k = 1 k jk Squared Eucldean dstance Assumpton: All features

More information

Cluster Analysis of Electrical Behavior

Cluster Analysis of Electrical Behavior Journal of Computer and Communcatons, 205, 3, 88-93 Publshed Onlne May 205 n ScRes. http://www.scrp.org/ournal/cc http://dx.do.org/0.4236/cc.205.350 Cluster Analyss of Electrcal Behavor Ln Lu Ln Lu, School

More information

Module Management Tool in Software Development Organizations

Module Management Tool in Software Development Organizations Journal of Computer Scence (5): 8-, 7 ISSN 59-66 7 Scence Publcatons Management Tool n Software Development Organzatons Ahmad A. Al-Rababah and Mohammad A. Al-Rababah Faculty of IT, Al-Ahlyyah Amman Unversty,

More information

Comparison of Heuristics for Scheduling Independent Tasks on Heterogeneous Distributed Environments

Comparison of Heuristics for Scheduling Independent Tasks on Heterogeneous Distributed Environments Comparson of Heurstcs for Schedulng Independent Tasks on Heterogeneous Dstrbuted Envronments Hesam Izakan¹, Ath Abraham², Senor Member, IEEE, Václav Snášel³ ¹ Islamc Azad Unversty, Ramsar Branch, Ramsar,

More information

Mathematics 256 a course in differential equations for engineering students

Mathematics 256 a course in differential equations for engineering students Mathematcs 56 a course n dfferental equatons for engneerng students Chapter 5. More effcent methods of numercal soluton Euler s method s qute neffcent. Because the error s essentally proportonal to the

More information

Private Information Retrieval (PIR)

Private Information Retrieval (PIR) 2 Levente Buttyán Problem formulaton Alce wants to obtan nformaton from a database, but she does not want the database to learn whch nformaton she wanted e.g., Alce s an nvestor queryng a stock-market

More information

An Entropy-Based Approach to Integrated Information Needs Assessment

An Entropy-Based Approach to Integrated Information Needs Assessment Dstrbuton Statement A: Approved for publc release; dstrbuton s unlmted. An Entropy-Based Approach to ntegrated nformaton Needs Assessment June 8, 2004 Wllam J. Farrell Lockheed Martn Advanced Technology

More information

CS 534: Computer Vision Model Fitting

CS 534: Computer Vision Model Fitting CS 534: Computer Vson Model Fttng Sprng 004 Ahmed Elgammal Dept of Computer Scence CS 534 Model Fttng - 1 Outlnes Model fttng s mportant Least-squares fttng Maxmum lkelhood estmaton MAP estmaton Robust

More information

Reliability and Energy-aware Cache Reconfiguration for Embedded Systems

Reliability and Energy-aware Cache Reconfiguration for Embedded Systems Relablty and Energy-aware Cache Reconfguraton for Embedded Systems Yuanwen Huang and Prabhat Mshra Department of Computer and Informaton Scence and Engneerng Unversty of Florda, Ganesvlle FL 326-62, USA

More information

Multitasking and Real-time Scheduling

Multitasking and Real-time Scheduling Multtaskng and Real-tme Schedulng EE8205: Embedded Computer Systems http://www.ee.ryerson.ca/~courses/ee8205/ Dr. Gul N. Khan http://www.ee.ryerson.ca/~gnkhan Electrcal and Computer Engneerng Ryerson Unversty

More information

Kent State University CS 4/ Design and Analysis of Algorithms. Dept. of Math & Computer Science LECT-16. Dynamic Programming

Kent State University CS 4/ Design and Analysis of Algorithms. Dept. of Math & Computer Science LECT-16. Dynamic Programming CS 4/560 Desgn and Analyss of Algorthms Kent State Unversty Dept. of Math & Computer Scence LECT-6 Dynamc Programmng 2 Dynamc Programmng Dynamc Programmng, lke the dvde-and-conquer method, solves problems

More information

5 The Primal-Dual Method

5 The Primal-Dual Method 5 The Prmal-Dual Method Orgnally desgned as a method for solvng lnear programs, where t reduces weghted optmzaton problems to smpler combnatoral ones, the prmal-dual method (PDM) has receved much attenton

More information

Real-Time Systems. Real-Time Systems. Verification by testing. Verification by testing

Real-Time Systems. Real-Time Systems. Verification by testing. Verification by testing EDA222/DIT161 Real-Tme Systems, Chalmers/GU, 2014/2015 Lecture #8 Real-Tme Systems Real-Tme Systems Lecture #8 Specfcaton Professor Jan Jonsson Implementaton System models Executon-tme analyss Department

More information

Space-Optimal, Wait-Free Real-Time Synchronization

Space-Optimal, Wait-Free Real-Time Synchronization 1 Space-Optmal, Wat-Free Real-Tme Synchronzaton Hyeonjoong Cho, Bnoy Ravndran ECE Dept., Vrgna Tech Blacksburg, VA 24061, USA {hjcho,bnoy}@vt.edu E. Douglas Jensen The MITRE Corporaton Bedford, MA 01730,

More information

Virtual Memory. Background. No. 10. Virtual Memory: concept. Logical Memory Space (review) Demand Paging(1) Virtual Memory

Virtual Memory. Background. No. 10. Virtual Memory: concept. Logical Memory Space (review) Demand Paging(1) Virtual Memory Background EECS. Operatng System Fundamentals No. Vrtual Memory Prof. Hu Jang Department of Electrcal Engneerng and Computer Scence, York Unversty Memory-management methods normally requres the entre process

More information

An Optimal Bandwidth Allocation and Data Droppage Scheme for Differentiated Services in a Wireless Network

An Optimal Bandwidth Allocation and Data Droppage Scheme for Differentiated Services in a Wireless Network Purdue Unversty Purdue e-pubs ECE Techncal Reports Electrcal and Computer Engneerng 3--7 An Optmal Bandwdth Allocaton and Data Droppage Scheme for Dfferentated Servces n a Wreless Network Waseem Shekh

More information

Load Balancing for Hex-Cell Interconnection Network

Load Balancing for Hex-Cell Interconnection Network Int. J. Communcatons, Network and System Scences,,, - Publshed Onlne Aprl n ScRes. http://www.scrp.org/journal/jcns http://dx.do.org/./jcns.. Load Balancng for Hex-Cell Interconnecton Network Saher Manaseer,

More information

For instance, ; the five basic number-sets are increasingly more n A B & B A A = B (1)

For instance, ; the five basic number-sets are increasingly more n A B & B A A = B (1) Secton 1.2 Subsets and the Boolean operatons on sets If every element of the set A s an element of the set B, we say that A s a subset of B, or that A s contaned n B, or that B contans A, and we wrte A

More information

A MOVING MESH APPROACH FOR SIMULATION BUDGET ALLOCATION ON CONTINUOUS DOMAINS

A MOVING MESH APPROACH FOR SIMULATION BUDGET ALLOCATION ON CONTINUOUS DOMAINS Proceedngs of the Wnter Smulaton Conference M E Kuhl, N M Steger, F B Armstrong, and J A Jones, eds A MOVING MESH APPROACH FOR SIMULATION BUDGET ALLOCATION ON CONTINUOUS DOMAINS Mark W Brantley Chun-Hung

More information

Analysis of Continuous Beams in General

Analysis of Continuous Beams in General Analyss of Contnuous Beams n General Contnuous beams consdered here are prsmatc, rgdly connected to each beam segment and supported at varous ponts along the beam. onts are selected at ponts of support,

More information

Today s Outline. Sorting: The Big Picture. Why Sort? Selection Sort: Idea. Insertion Sort: Idea. Sorting Chapter 7 in Weiss.

Today s Outline. Sorting: The Big Picture. Why Sort? Selection Sort: Idea. Insertion Sort: Idea. Sorting Chapter 7 in Weiss. Today s Outlne Sortng Chapter 7 n Wess CSE 26 Data Structures Ruth Anderson Announcements Wrtten Homework #6 due Frday 2/26 at the begnnng of lecture Proect Code due Mon March 1 by 11pm Today s Topcs:

More information

Course Introduction. Algorithm 8/31/2017. COSC 320 Advanced Data Structures and Algorithms. COSC 320 Advanced Data Structures and Algorithms

Course Introduction. Algorithm 8/31/2017. COSC 320 Advanced Data Structures and Algorithms. COSC 320 Advanced Data Structures and Algorithms Course Introducton Course Topcs Exams, abs, Proects A quc loo at a few algorthms 1 Advanced Data Structures and Algorthms Descrpton: We are gong to dscuss algorthm complexty analyss, algorthm desgn technques

More information

A protocol for mixed-criticality management in switched Ethernet networks

A protocol for mixed-criticality management in switched Ethernet networks A protocol for mxed-crtcalty management n swtched Ethernet networks Olver CROS, Laurent GEORGE Unversté Pars-Est, LIGM / ESIEE, France cros@ece.fr,lgeorge@eee.org Xaotng LI ECE Pars / LACSC, France xaotng.l@ece.fr

More information

Repeater Insertion for Two-Terminal Nets in Three-Dimensional Integrated Circuits

Repeater Insertion for Two-Terminal Nets in Three-Dimensional Integrated Circuits Repeater Inserton for Two-Termnal Nets n Three-Dmensonal Integrated Crcuts Hu Xu, Vasls F. Pavlds, and Govann De Mchel LSI - EPFL, CH-5, Swtzerland, {hu.xu,vasleos.pavlds,govann.demchel}@epfl.ch Abstract.

More information

Chapter 1. Introduction

Chapter 1. Introduction Chapter 1 Introducton 1.1 Parallel Processng There s a contnual demand for greater computatonal speed from a computer system than s currently possble (.e. sequental systems). Areas need great computatonal

More information

Video Proxy System for a Large-scale VOD System (DINA)

Video Proxy System for a Large-scale VOD System (DINA) Vdeo Proxy System for a Large-scale VOD System (DINA) KWUN-CHUNG CHAN #, KWOK-WAI CHEUNG *# #Department of Informaton Engneerng *Centre of Innovaton and Technology The Chnese Unversty of Hong Kong SHATIN,

More information

Sum of Linear and Fractional Multiobjective Programming Problem under Fuzzy Rules Constraints

Sum of Linear and Fractional Multiobjective Programming Problem under Fuzzy Rules Constraints Australan Journal of Basc and Appled Scences, 2(4): 1204-1208, 2008 ISSN 1991-8178 Sum of Lnear and Fractonal Multobjectve Programmng Problem under Fuzzy Rules Constrants 1 2 Sanjay Jan and Kalash Lachhwan

More information

Performance Evaluation of Information Retrieval Systems

Performance Evaluation of Information Retrieval Systems Why System Evaluaton? Performance Evaluaton of Informaton Retreval Systems Many sldes n ths secton are adapted from Prof. Joydeep Ghosh (UT ECE) who n turn adapted them from Prof. Dk Lee (Unv. of Scence

More information

Efficient Distributed File System (EDFS)

Efficient Distributed File System (EDFS) Effcent Dstrbuted Fle System (EDFS) (Sem-Centralzed) Debessay(Debsh) Fesehaye, Rahul Malk & Klara Naherstedt Unversty of Illnos-Urbana Champagn Contents Problem Statement, Related Work, EDFS Desgn Rate

More information

On the Fairness-Efficiency Tradeoff for Packet Processing with Multiple Resources

On the Fairness-Efficiency Tradeoff for Packet Processing with Multiple Resources On the Farness-Effcency Tradeoff for Packet Processng wth Multple Resources We Wang, Chen Feng, Baochun L, and Ben Lang Department of Electrcal and Computer Engneerng, Unversty of Toronto {wewang, cfeng,

More information

Solving two-person zero-sum game by Matlab

Solving two-person zero-sum game by Matlab Appled Mechancs and Materals Onlne: 2011-02-02 ISSN: 1662-7482, Vols. 50-51, pp 262-265 do:10.4028/www.scentfc.net/amm.50-51.262 2011 Trans Tech Publcatons, Swtzerland Solvng two-person zero-sum game by

More information

CMPS 10 Introduction to Computer Science Lecture Notes

CMPS 10 Introduction to Computer Science Lecture Notes CPS 0 Introducton to Computer Scence Lecture Notes Chapter : Algorthm Desgn How should we present algorthms? Natural languages lke Englsh, Spansh, or French whch are rch n nterpretaton and meanng are not

More information

Real-Time Guarantees. Traffic Characteristics. Flow Control

Real-Time Guarantees. Traffic Characteristics. Flow Control Real-Tme Guarantees Requrements on RT communcaton protocols: delay (response s) small jtter small throughput hgh error detecton at recever (and sender) small error detecton latency no thrashng under peak

More information

Helsinki University Of Technology, Systems Analysis Laboratory Mat Independent research projects in applied mathematics (3 cr)

Helsinki University Of Technology, Systems Analysis Laboratory Mat Independent research projects in applied mathematics (3 cr) Helsnk Unversty Of Technology, Systems Analyss Laboratory Mat-2.08 Independent research projects n appled mathematcs (3 cr) "! #$&% Antt Laukkanen 506 R ajlaukka@cc.hut.f 2 Introducton...3 2 Multattrbute

More information

3. CR parameters and Multi-Objective Fitness Function

3. CR parameters and Multi-Objective Fitness Function 3 CR parameters and Mult-objectve Ftness Functon 41 3. CR parameters and Mult-Objectve Ftness Functon 3.1. Introducton Cogntve rados dynamcally confgure the wreless communcaton system, whch takes beneft

More information

S1 Note. Basis functions.

S1 Note. Basis functions. S1 Note. Bass functons. Contents Types of bass functons...1 The Fourer bass...2 B-splne bass...3 Power and type I error rates wth dfferent numbers of bass functons...4 Table S1. Smulaton results of type

More information

Analysis of Collaborative Distributed Admission Control in x Networks

Analysis of Collaborative Distributed Admission Control in x Networks 1 Analyss of Collaboratve Dstrbuted Admsson Control n 82.11x Networks Thnh Nguyen, Member, IEEE, Ken Nguyen, Member, IEEE, Lnha He, Member, IEEE, Abstract Wth the recent surge of wreless home networks,

More information

A Saturation Binary Neural Network for Crossbar Switching Problem

A Saturation Binary Neural Network for Crossbar Switching Problem A Saturaton Bnary Neural Network for Crossbar Swtchng Problem Cu Zhang 1, L-Qng Zhao 2, and Rong-Long Wang 2 1 Department of Autocontrol, Laonng Insttute of Scence and Technology, Benx, Chna bxlkyzhangcu@163.com

More information

An Application of the Dulmage-Mendelsohn Decomposition to Sparse Null Space Bases of Full Row Rank Matrices

An Application of the Dulmage-Mendelsohn Decomposition to Sparse Null Space Bases of Full Row Rank Matrices Internatonal Mathematcal Forum, Vol 7, 2012, no 52, 2549-2554 An Applcaton of the Dulmage-Mendelsohn Decomposton to Sparse Null Space Bases of Full Row Rank Matrces Mostafa Khorramzadeh Department of Mathematcal

More information

Channel 0. Channel 1 Channel 2. Channel 3 Channel 4. Channel 5 Channel 6 Channel 7

Channel 0. Channel 1 Channel 2. Channel 3 Channel 4. Channel 5 Channel 6 Channel 7 Optmzed Regonal Cachng for On-Demand Data Delvery Derek L. Eager Mchael C. Ferrs Mary K. Vernon Unversty of Saskatchewan Unversty of Wsconsn Madson Saskatoon, SK Canada S7N 5A9 Madson, WI 5376 eager@cs.usask.ca

More information

ELEC 377 Operating Systems. Week 6 Class 3

ELEC 377 Operating Systems. Week 6 Class 3 ELEC 377 Operatng Systems Week 6 Class 3 Last Class Memory Management Memory Pagng Pagng Structure ELEC 377 Operatng Systems Today Pagng Szes Vrtual Memory Concept Demand Pagng ELEC 377 Operatng Systems

More information

CSE 326: Data Structures Quicksort Comparison Sorting Bound

CSE 326: Data Structures Quicksort Comparison Sorting Bound CSE 326: Data Structures Qucksort Comparson Sortng Bound Steve Setz Wnter 2009 Qucksort Qucksort uses a dvde and conquer strategy, but does not requre the O(N) extra space that MergeSort does. Here s the

More information

A Binarization Algorithm specialized on Document Images and Photos

A Binarization Algorithm specialized on Document Images and Photos A Bnarzaton Algorthm specalzed on Document mages and Photos Ergna Kavalleratou Dept. of nformaton and Communcaton Systems Engneerng Unversty of the Aegean kavalleratou@aegean.gr Abstract n ths paper, a

More information

CS 268: Lecture 8 Router Support for Congestion Control

CS 268: Lecture 8 Router Support for Congestion Control CS 268: Lecture 8 Router Support for Congeston Control Ion Stoca Computer Scence Dvson Department of Electrcal Engneerng and Computer Scences Unversty of Calforna, Berkeley Berkeley, CA 9472-1776 Router

More information

On Achieving Fairness in the Joint Allocation of Buffer and Bandwidth Resources: Principles and Algorithms

On Achieving Fairness in the Joint Allocation of Buffer and Bandwidth Resources: Principles and Algorithms On Achevng Farness n the Jont Allocaton of Buffer and Bandwdth Resources: Prncples and Algorthms Yunka Zhou and Harsh Sethu (correspondng author) Abstract Farness n network traffc management can mprove

More information

A Fast Content-Based Multimedia Retrieval Technique Using Compressed Data

A Fast Content-Based Multimedia Retrieval Technique Using Compressed Data A Fast Content-Based Multmeda Retreval Technque Usng Compressed Data Borko Furht and Pornvt Saksobhavvat NSF Multmeda Laboratory Florda Atlantc Unversty, Boca Raton, Florda 3343 ABSTRACT In ths paper,

More information

DESIGNING TRANSMISSION SCHEDULES FOR WIRELESS AD HOC NETWORKS TO MAXIMIZE NETWORK THROUGHPUT

DESIGNING TRANSMISSION SCHEDULES FOR WIRELESS AD HOC NETWORKS TO MAXIMIZE NETWORK THROUGHPUT DESIGNING TRANSMISSION SCHEDULES FOR WIRELESS AD HOC NETWORKS TO MAXIMIZE NETWORK THROUGHPUT Bran J. Wolf, Joseph L. Hammond, and Harlan B. Russell Dept. of Electrcal and Computer Engneerng, Clemson Unversty,

More information

Multi-objective Virtual Machine Placement for Load Balancing

Multi-objective Virtual Machine Placement for Load Balancing Mult-obectve Vrtual Machne Placement for Load Balancng Feng FANG and Bn-Bn Qu,a School of Computer Scence & Technology, Huazhong Unversty Of Scence And Technology, Wuhan, Chna Abstract. The vrtual machne

More information

A Predictable Execution Model for COTS-based Embedded Systems

A Predictable Execution Model for COTS-based Embedded Systems 2011 17th IEEE Real-Tme and Embedded Technology and Applcatons Symposum A Predctable Executon Model for COTS-based Embedded Systems Rodolfo Pellzzon, Emlano Bett, Stanley Bak, Gang Yao, John Crswell, Marco

More information

A SYSTOLIC APPROACH TO LOOP PARTITIONING AND MAPPING INTO FIXED SIZE DISTRIBUTED MEMORY ARCHITECTURES

A SYSTOLIC APPROACH TO LOOP PARTITIONING AND MAPPING INTO FIXED SIZE DISTRIBUTED MEMORY ARCHITECTURES A SYSOLIC APPROACH O LOOP PARIIONING AND MAPPING INO FIXED SIZE DISRIBUED MEMORY ARCHIECURES Ioanns Drosts, Nektaros Kozrs, George Papakonstantnou and Panayots sanakas Natonal echncal Unversty of Athens

More information

Scheduling Remote Access to Scientific Instruments in Cyberinfrastructure for Education and Research

Scheduling Remote Access to Scientific Instruments in Cyberinfrastructure for Education and Research Schedulng Remote Access to Scentfc Instruments n Cybernfrastructure for Educaton and Research Je Yn 1, Junwe Cao 2,3,*, Yuexuan Wang 4, Lanchen Lu 1,3 and Cheng Wu 1,3 1 Natonal CIMS Engneerng and Research

More information

Load-Balanced Anycast Routing

Load-Balanced Anycast Routing Load-Balanced Anycast Routng Chng-Yu Ln, Jung-Hua Lo, and Sy-Yen Kuo Department of Electrcal Engneerng atonal Tawan Unversty, Tape, Tawan sykuo@cc.ee.ntu.edu.tw Abstract For fault-tolerance and load-balance

More information

Simulation Based Analysis of FAST TCP using OMNET++

Simulation Based Analysis of FAST TCP using OMNET++ Smulaton Based Analyss of FAST TCP usng OMNET++ Umar ul Hassan 04030038@lums.edu.pk Md Term Report CS678 Topcs n Internet Research Sprng, 2006 Introducton Internet traffc s doublng roughly every 3 months

More information

A Frame Packing Mechanism Using PDO Communication Service within CANopen

A Frame Packing Mechanism Using PDO Communication Service within CANopen 28 A Frame Packng Mechansm Usng PDO Communcaton Servce wthn CANopen Mnkoo Kang and Kejn Park Dvson of Industral & Informaton Systems Engneerng, Ajou Unversty, Suwon, Gyeongg-do, South Korea Summary The

More information

Tolerating Transient Faults in Statically Scheduled Safety-Critical Embedded Systems

Tolerating Transient Faults in Statically Scheduled Safety-Critical Embedded Systems Toleratng Transent Faults n Statcally Scheduled Safety-Crtcal Embedded Systems Nagaraan Kandasamy *, John P. Hayes *, and Bran T. Murray ** * Department of Electrcal Engneerng ** Delph Automotve Systems

More information

CSE 326: Data Structures Quicksort Comparison Sorting Bound

CSE 326: Data Structures Quicksort Comparison Sorting Bound CSE 326: Data Structures Qucksort Comparson Sortng Bound Bran Curless Sprng 2008 Announcements (5/14/08) Homework due at begnnng of class on Frday. Secton tomorrow: Graded homeworks returned More dscusson

More information

A Hybrid Genetic Algorithm for Routing Optimization in IP Networks Utilizing Bandwidth and Delay Metrics

A Hybrid Genetic Algorithm for Routing Optimization in IP Networks Utilizing Bandwidth and Delay Metrics A Hybrd Genetc Algorthm for Routng Optmzaton n IP Networks Utlzng Bandwdth and Delay Metrcs Anton Redl Insttute of Communcaton Networks, Munch Unversty of Technology, Arcsstr. 21, 80290 Munch, Germany

More information

Determining the Optimal Bandwidth Based on Multi-criterion Fusion

Determining the Optimal Bandwidth Based on Multi-criterion Fusion Proceedngs of 01 4th Internatonal Conference on Machne Learnng and Computng IPCSIT vol. 5 (01) (01) IACSIT Press, Sngapore Determnng the Optmal Bandwdth Based on Mult-crteron Fuson Ha-L Lang 1+, Xan-Mn

More information

Brave New World Pseudocode Reference

Brave New World Pseudocode Reference Brave New World Pseudocode Reference Pseudocode s a way to descrbe how to accomplsh tasks usng basc steps lke those a computer mght perform. In ths week s lab, you'll see how a form of pseudocode can be

More information

Fibre-Optic AWG-based Real-Time Networks

Fibre-Optic AWG-based Real-Time Networks Fbre-Optc AWG-based Real-Tme Networks Krstna Kunert, Annette Böhm, Magnus Jonsson, School of Informaton Scence, Computer and Electrcal Engneerng, Halmstad Unversty {Magnus.Jonsson, Krstna.Kunert}@de.hh.se

More information

Quantifying Responsiveness of TCP Aggregates by Using Direct Sequence Spread Spectrum CDMA and Its Application in Congestion Control

Quantifying Responsiveness of TCP Aggregates by Using Direct Sequence Spread Spectrum CDMA and Its Application in Congestion Control Quantfyng Responsveness of TCP Aggregates by Usng Drect Sequence Spread Spectrum CDMA and Its Applcaton n Congeston Control Mehd Kalantar Department of Electrcal and Computer Engneerng Unversty of Maryland,

More information

Wishing you all a Total Quality New Year!

Wishing you all a Total Quality New Year! Total Qualty Management and Sx Sgma Post Graduate Program 214-15 Sesson 4 Vnay Kumar Kalakband Assstant Professor Operatons & Systems Area 1 Wshng you all a Total Qualty New Year! Hope you acheve Sx sgma

More information

Complex Numbers. Now we also saw that if a and b were both positive then ab = a b. For a second let s forget that restriction and do the following.

Complex Numbers. Now we also saw that if a and b were both positive then ab = a b. For a second let s forget that restriction and do the following. Complex Numbers The last topc n ths secton s not really related to most of what we ve done n ths chapter, although t s somewhat related to the radcals secton as we wll see. We also won t need the materal

More information

Routing in Degree-constrained FSO Mesh Networks

Routing in Degree-constrained FSO Mesh Networks Internatonal Journal of Hybrd Informaton Technology Vol., No., Aprl, 009 Routng n Degree-constraned FSO Mesh Networks Zpng Hu, Pramode Verma, and James Sluss Jr. School of Electrcal & Computer Engneerng

More information

CHAPTER 2 DECOMPOSITION OF GRAPHS

CHAPTER 2 DECOMPOSITION OF GRAPHS CHAPTER DECOMPOSITION OF GRAPHS. INTRODUCTION A graph H s called a Supersubdvson of a graph G f H s obtaned from G by replacng every edge uv of G by a bpartte graph,m (m may vary for each edge by dentfyng

More information

WIRELESS communication technology has gained widespread

WIRELESS communication technology has gained widespread 616 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 4, NO. 6, NOVEMBER/DECEMBER 2005 Dstrbuted Far Schedulng n a Wreless LAN Ntn Vadya, Senor Member, IEEE, Anurag Dugar, Seema Gupta, and Paramvr Bahl, Senor

More information

Report on On-line Graph Coloring

Report on On-line Graph Coloring 2003 Fall Semester Comp 670K Onlne Algorthm Report on LO Yuet Me (00086365) cndylo@ust.hk Abstract Onlne algorthm deals wth data that has no future nformaton. Lots of examples demonstrate that onlne algorthm

More information

Some material adapted from Mohamed Younis, UMBC CMSC 611 Spr 2003 course slides Some material adapted from Hennessy & Patterson / 2003 Elsevier

Some material adapted from Mohamed Younis, UMBC CMSC 611 Spr 2003 course slides Some material adapted from Hennessy & Patterson / 2003 Elsevier Some materal adapted from Mohamed Youns, UMBC CMSC 611 Spr 2003 course sldes Some materal adapted from Hennessy & Patterson / 2003 Elsever Scence Performance = 1 Executon tme Speedup = Performance (B)

More information

Fast Computation of Shortest Path for Visiting Segments in the Plane

Fast Computation of Shortest Path for Visiting Segments in the Plane Send Orders for Reprnts to reprnts@benthamscence.ae 4 The Open Cybernetcs & Systemcs Journal, 04, 8, 4-9 Open Access Fast Computaton of Shortest Path for Vstng Segments n the Plane Ljuan Wang,, Bo Jang

More information

2x x l. Module 3: Element Properties Lecture 4: Lagrange and Serendipity Elements

2x x l. Module 3: Element Properties Lecture 4: Lagrange and Serendipity Elements Module 3: Element Propertes Lecture : Lagrange and Serendpty Elements 5 In last lecture note, the nterpolaton functons are derved on the bass of assumed polynomal from Pascal s trangle for the fled varable.

More information

Pricing Network Resources for Adaptive Applications in a Differentiated Services Network

Pricing Network Resources for Adaptive Applications in a Differentiated Services Network IEEE INFOCOM Prcng Network Resources for Adaptve Applcatons n a Dfferentated Servces Network Xn Wang and Hennng Schulzrnne Columba Unversty Emal: {xnwang, schulzrnne}@cs.columba.edu Abstract The Dfferentated

More information

WITH rapid improvements of wireless technologies,

WITH rapid improvements of wireless technologies, JOURNAL OF SYSTEMS ARCHITECTURE, SPECIAL ISSUE: HIGHLY-RELIABLE CPS, VOL. 00, NO. 0, MONTH 013 1 Adaptve GTS Allocaton n IEEE 80.15.4 for Real-Tme Wreless Sensor Networks Feng Xa, Ruonan Hao, Je L, Naxue

More information

Real-time Fault-tolerant Scheduling Algorithm for Distributed Computing Systems

Real-time Fault-tolerant Scheduling Algorithm for Distributed Computing Systems Real-tme Fault-tolerant Schedulng Algorthm for Dstrbuted Computng Systems Yun Lng, Y Ouyang College of Computer Scence and Informaton Engneerng Zheang Gongshang Unversty Postal code: 310018 P.R.CHINA {ylng,

More information

Configuration Management in Multi-Context Reconfigurable Systems for Simultaneous Performance and Power Optimizations*

Configuration Management in Multi-Context Reconfigurable Systems for Simultaneous Performance and Power Optimizations* Confguraton Management n Mult-Context Reconfgurable Systems for Smultaneous Performance and Power Optmzatons* Rafael Maestre, Mlagros Fernandez Departamento de Arqutectura de Computadores y Automátca Unversdad

More information

Hermite Splines in Lie Groups as Products of Geodesics

Hermite Splines in Lie Groups as Products of Geodesics Hermte Splnes n Le Groups as Products of Geodescs Ethan Eade Updated May 28, 2017 1 Introducton 1.1 Goal Ths document defnes a curve n the Le group G parametrzed by tme and by structural parameters n the

More information