Shared Running Buffer Based Proxy Caching of Streaming Sessions

Size: px
Start display at page:

Download "Shared Running Buffer Based Proxy Caching of Streaming Sessions"

Transcription

1 Shared Runnng Buffer Based Proxy Cachng of Streamng Sessons Songqng Chen, Bo Shen, Yong Yan, Sujoy Basu Moble and Meda Systems Laboratory HP Laboratores Palo Alto HPL March th, 23* E-mal: {boshen, yongyan, shared runnng buffer, proxy cachng, patchng, streamng meda delvery, VOD Wth the fallng prce of memory, an ncreasng number of multmeda servers and proxes are now equpped wth a large memory space. Cachng meda objects n memory of a proxy helps to reduce the network traffc, the dsk I/O bandwdth requrement, and the data delvery latency. The runnng buffer approach and ts alternatves are representatve technques to cachng streamng data n the memory. There are two lmts n the exstng technques. Frst, although multple runnng buffers for the same meda object co-exst n a gven processng perod, data sharng among multple buffers s not consdered. Second, user access patterns are not nsghtfully consdered n the buffer management. In ths paper, we propose two technques based on shared runnng buffers (SRB) n the proxy to address these lmts. Consderng user access patterns and characterstcs of the requested meda objects, our technques adaptvely allocate memory buffers to fully utlze the currently buffered data of streamng sessons, wth the am to reduce both the server load and the network traffc. Expermentally comparng wth several exstng technques, we show that the proposed technques acheve sgnfcant performance mprovement by effectvely usng the shared runnng buffers. * Internal Accesson Date Only Approved for External Publcaton Department of Computer Scence, College of Wllam and Mary, Wllamsburg, VA 2388 Copyrght Hewlett-Packard Company 23

2 Shared Runnng Buffer Based Proxy Cachng of Streamng Sessons Songqng Chen Λ Bo Shen, Yong Yan and Sujoy Basu Department of Computer Scence Moble and Meda System Lab College of Wllam and Mary Page Mll Road Wllamsburg, VA 2388 Palo Alto, CA Abstract Wth the fallng prce of memory, an ncreasng number of multmeda servers and proxes are now equpped wth a large memory space. Cachng meda objects n memory of a proxy helps to reduce the network traffc, the dsk I/O bandwdth requrement, and the data delvery latency. The runnng buffer approach and ts alternatves are representatve technques to cachng streamng data n the memory. There are two lmts n the exstng technques. Frst, although multple runnng buffers for the same meda object co-exst n a gven processng perod, data sharng among multple buffers s not consdered. Second, user access patterns are not nsghtfully consdered n the buffer management. In ths paper, we propose two technques based on shared runnng buffers (SRB) n the proxy to address these lmts. Consderng user access patterns and characterstcs of the requested meda objects, our technques adaptvely allocate memory buffers to fully utlze the currently buffered data of streamng sessons, wth the am to reduce both the server load and the network traffc. Expermentally comparng wth several exstng technques, we show that the proposed technques acheve sgnfcant performance mprovement by effectvely usng the shared runnng buffers. Key Words: Shared Runnng Buffer, Proxy Cachng, Patchng, Streamng Meda, VOD. Introducton The buldng block of a content delvery network s a server-proxy-clent system. In ths system, the server delvers the content to the clent through a proxy. The proxy can choose to cache the object so that subsequent requests to the same object can be served drectly from the proxy wthout contactng the server. Proxy cachng strateges have therefore been the focus of many studes. Much work has been done n cachng statc web content to reduce network load and end-to-end latency. Typcal examples of such work nclude CERN httpd [6], Harvest [4] and Squd [4]. The cachng of streamng meda content presents a dfferent set of challenges: () The sze of a streamng meda object s usually orders of magntudes larger than tradtonal web contents. For Λ Work performed as an ntern at HP Labs, summer 22.

3 example, a two-hour long MPEG vdeo requres about.4 GB of dsk space, whle tradtonal web objects are of the magntude of KB; () The demand of contnuous and tmely delvery of a streamng meda object s more rgorous than that of text-based Web pages. Therefore a lot of resources have to be reserved for delverng the streamng meda data to a clent. In practce, even a relatvely small number of clents can overload a meda server, creatng bottlenecks by demandng hgh dsk bandwdth on the server and hght network bandwdth to the clents. To address these challenges, researchers have proposed dfferent methods to cache streamng meda objects va partal cachng, patchng or proxy bufferng. In the partal cachng approach, ether a prefx [8] or segments [2] of a meda object nstead of the whole object s/are cached. Therefore, less storage space s requred. For on-gong streamng sessons, patchng can be used so that later sessons for the same object can be served smultaneously. For proxy bufferng, ether a fxed-sze runnng buffer [3] or an nterval [9] can be used to allocate buffers to buffer a runnng wndow ofan on-gong streamng sesson n the memory of the proxy. Among these technques, partal cachng uses dsk resource on the proxy; patchng uses storage resource on the clent sde, and theoretcally no memory resource s requred at the proxy; proxy bufferng uses the memory resource on the proxy. However, nether runnng buffer nor nterval cachng uses the memory resource to the full extent. More detaled analyss of these technques can be found n Secton 2. In ths paper, we frst propose a new memory-based cachng algorthm for streamng meda objects usng Shared Runnng Buffers (SRB). In ths algorthm, the memory space on the proxy s allocated adaptvely based on the user access pattern and the requested meda objects themselves. Streamng sessons are cached n runnng buffers. Ths algorthm dynamcally caches meda objects n the memory whle delverng the data to the clent so that the bottleneck of the dsk and/or network I/O s releved. More mportantly, smlar sessons can share dfferent runs of the on-gong sessons. Ths approach s especally useful when requests to streamng objects are hghly temporal localzed. The SRB algorthm () adaptvely allocates memory space accordng to the user access pattern; () enables maxmal sharng of the cached data n memory; () optmally reclams memory space when requests termnate; (v) apples a near-optmal replacement polcy n the real tme. Based on the SRB meda cachng algorthm, we further propose an effcent meda delverng algorthm: (PSRB), whch further mproves the performance of the meda data delvery wthout the necessty of cachng. Smulatons are conducted based on synthetc workloads of web meda objects wth mxed lengths as well as workloads wth accesses to lengthy meda objects n the VOD envronment. In addton, we use an access log of a meda server n a real enterprse ntranet to further conduct smulatons. The smulaton results ndcate that the performance of our algorthms s superor to prevous solutons. The rest of the paper s organzed as follows. In Secton 2, the related work s surveyed. Secton 3 descrbes the optmal memory-based cachng algorthms we propose. To test the performance of the proposed algorthms, we use synthetc workloads as well as real workload from an enterprse meda server. Some statstcal analyss of these workloads s provded n Secton 4. Performance evaluatons are conducted n Secton. We then make concludng remarks n Secton 6. 2 Related Work In ths secton, we survey prevous work related to the cachng of streamng meda content. Three types of methods are nvestgated. Frst, we brefly dscuss streamng meda cachng methods that 2

4 use the storage resource on the proxy. These methods cache meda objects for streamng sessons that are not overlapped n tme. The cached data perssts n cache even after sesson termnaton. Usually, a long-term storage resource such as dsk storage s used. Second, we nvestgate cachng methods that use the storage resource on the clent devce. These methods share streamng sessons that are overlapped n tme. However, the proxy does not buffer data for any sesson. Lastly, we study cachng methods that use the storage resource on the proxy to buffer portons of meda objects for streamng sessons that are overlapped n tme. Note that the buffered data perssts n cache only when the sessons are alve. Therefore, short-term storage resource such as RAM s used. 2. Partal Cachng Storng the entre meda object n the proxy may be neffcent f mostly small portons of very large meda objects are accessed. Ths s partcularly true f the cached streamng meda object s not popular. The frst ntuton s to cache portons of the meda object. These partal cachng systems always use the storage on the proxy. Some early work on the storage for meda objects can be found n [2, 9]. Two typcal types of partal cachng have been nvestgated. Prefx cachng [8] stores only the frst part (prefx) of the popular meda object. When a clent requests a meda stream whose prefx s cached, the proxy delvers the prefx to the clent whle t requests the remander of the object from the orgn server. By cachng a (large enough) prefx, the start-up latency perceved by the clent s reduced. The challenge les n the determnaton of the prefx sze. Items such as roundtrp delay, server-to-proxy latency, vdeo specfc parameters (e.g., sze, bt rate, etc.), and retransmsson rate of lost packets can be consdered n calculatng the approprate prefx length. Alternatvely, meda objects can be cached n segments. Ths s partcularly useful when clents only vew portons of the objects. The segments n whch clents are not nterested wll not be cached. Wu et al. [2] use segments wth exponentally ncremental sze to model the fact that clents usually start vewng a streamng meda object from the begnnng and are more and more lkely to termnate the sesson toward the end of the object. A combnaton of fxed length and exponental length segment-based cachng method s consdered n the RCache and Slo project []. Lee et al. [] uses a context-aware segmentaton so that segments of user nterest are cached. 2.2 Sesson Sharng The delvery of a streamng meda object takes tme to complete. We call ths delvery process a streamng sesson. Sharng s possble among sessons that overlap. To ths end, varous knds of patchng schemes are proposed for sharng along the tme lne. Ths type of work s typcally seen n research related to vdeo on demand (VOD) systems. Patchng algorthms [3] use storage on the clent devce so that a clent can lsten to multple channels. Greedy patchng always patches to the exstng full streamng sesson whle grace patchng restarts a new full streamng sesson at some approprate pont n tme. Optmal patchng [7] further assumes suffcent storage resource at the clent sde to receve as much data as possble whle lstenng to as many channels as possble. Fgure shows the basc deas of the patchng technques through examples. In these fgures, meda poston ndcates a tme poston at whch offset the meda object s delvered 3

5 Meda Poston Meda Poston Meda Poston Access Tme Access Tme Access Tme Fgure : Left to rght, (a) Greedy Patchng, (b) Grace Patchng, (c) Optmal Patchng. to the clent. For llustraton purpose, we assume nstantaneous delverng of meda content at a constant rate equal to the presentaton rate of the meda object. That s, f a request s served mmedately, the clent receves n tme unts of data after n tme unts snce the request s ntated. The sold lnes represent a streamng sesson between the server and the proxy. The goal of these algorthms s to reduce the server-to-proxy traffc. A greedy patchng example s shown n Fgure (a). A full server-to-proxy sesson s establshed at the frst request arrval for a certan meda object. All subsequent requests for the same meda object are served as patches to the ths full sesson before t termnates, at whch pont another full sesson s establshed for the next request arrval. It s obvous that the patchng sesson can be excessvely long as shown n the last two requests n Fgure (a). It s more effcent to start a new sesson even before the prevous full sesson termnates. The grace patchng example shown n Fgure (b) ndcates that upon the arrval of the ffth request, a new full server-to-proxy sesson s started. Therefore, the patchng sesson of the last request s sgnfcantly shorter than n the greedy patchng example. The optmal pont to start a new full sesson can be derved mathematcally gven certan request arrval pattern [2]. The optmal patchng scheme s ntroduced n [7]. The basc dea s that later sessons patch to as many on-gong sessons as possble. The on-gong sessons can be a full sesson or a patchng sesson. Fgure (c) shows an example scenaro. The last request patches to the patchng sesson started for the second last request as well as the full sesson started for the frst request. Ths approach acheves the maxmum server-to-proxy traffc reducton. On the other hand, t requres extra resource n clent storage and proxy-to-clent bandwdth as well as complex schedulng at the proxy. Note that no storage resource s requred on the proxy for these sesson sharng algorthms. On the other hand, snce the clents have to lsten to multple channels and store content before ts presentaton tme, clent sde storage s necessary. One specal type of the sesson-sharng approaches s the batchng approach [, ]. In ths approach, requests are grouped and served smultaneously va multcast. Therefore, requests arrvng earler have to wat. Hence, certan delay s ntroduced to the early arrved requests. 4

6 2.3 Proxy Bufferng To further mprove the cachng performance for streamng meda, memory resources on the proxy are used. Ths s more and more practcal gven that the prce for memory keeps fallng. For memory-based cachng, runnng buffer and nterval cachng methods have been studed. Meda Poston B B2 Access Tme R R2 R3 R4 R R6 (a) Meda Poston B R R2 R3 R4 R R6 (b) B2 Access Tme Fgure 2: (a) Runnng Buffer, (b) Interval Cachng. Dynamc cachng [3] uses a fxed-sze runnng buffer to cache a runnng wndow of a streamng sesson. It works as follows. When a request arrves, a buffer of a predetermned length s allocated to cache the meda data the proxy fetches from the server, n the expectaton that closely followed requests wll use the data n the memory nstead of fetchng from other sources (e.g., dsk or orgn server or other cooperatve caches). Fgure 2(a) shows an example of runnng buffer cachng. A fx-szed buffer s allocated upon the arrval of request R. The buffer s flled by the sesson started by R and run to the end of stream (the paralleled vertcal lnes ndcate the buffer fullness and the locaton of the buffered data relatve to the begnnng of the meda object). Subsequent requests R 2, R 3, R 4 are served from the buffer snce they arrved wthn the tme perod covered by the buffer. Request R arrves beyond the range covered by the frst buffer, hence a second buffer of the same predetermned sze s allocated. Subsequently, R 6 s served from the second buffer. On a dfferent approach, nterval cachng [8, 9] further consders request arrval patterns so that memory resource s more effcently used. Interval cachng consders closely arrved requests for the same meda object as a par and orders ther arrval ntervals globally. The subsequent allocaton of memory always favors smaller ntervals. Effectvely, more requests are served gven a fxed amount of memory. Fgure 2(b) shows an example of the nterval cachng method. Upon the arrval of request R 2, an nterval s formed, and a buffer of the sze equvalent to the nterval s allocated. The buffer s flled by the sesson started by R from ths pont on. The sesson ntated by R 2 only needs to receve the frst part of data from the server (the sold lne startng from R 2 ). It can receve the rest of data from the buffer. The same apples when R 3 and R 4 arrve. The stuaton changes untl the arrval of R. Snce the nterval between R 4 and R s smaller than that between R 3 and R 4, the buffer ntally allocated for the nterval of R 3 and R 4 (not flled up yet) s released, and the space s re-allocated for the new nterval of R 4 and R. Now the sesson for R 4 has to run to the end of the stream. Note that these two approaches use the memory resource on the proxy so that the clent s releved from the buffer requrement as n the patchng schemes dscussed n the prevous secton. In addton, one proxy-to-clent channel suffces for these bufferng schemes.

7 3 Shared Runnng Buffer (SRB) Meda Cachng Algorthm Bufferng streamng content n memory has been shown to have great potental to allevate contenton on the streamng server so that a larger number of sessons can be served. It has been shown that two exstng memory cachng approaches: runnng buffer cachng [3] and nterval cachng [8, 9], do not make effectve use of the lmted memory resource. Motvated by the lmts of the current memory bufferng approaches, we desgn a Shared Runnng Buffer (SRB) based cachng algorthm for streamng meda to better utlze memory. In ths secton, wth the ntroducton of several new concepts, we frst descrbe the basc SRB meda cachng algorthm n detal. Then, we present an extenson to the SRB: (PSRB). 3. Related Defntons The algorthm frst consders buffer allocaton n a tme span T startng from the frst request. We denote R j j as the j-th request to meda object, andt as the arrval tme of ths request. Assume that there are n request arrvals wthn the tme span T and R n s the last request arrved n T. For the convenence of representaton wthout losng precson, T s normalzed to and T j (where < j» n) s a tme relatve to T. Based on the above, the followng concepts are defned to capture the characterstcs of the user request pattern.. Interval Seres: An nterval s defned as the dfference n tme between two consecutve request arrvals. We denote I k as the k-th nterval for object. An Interval Seres conssts a group of ntervals. Wthn the tme T,fn =, the nterval I s defned as ; otherwse, t s defned as: 8 < I k = : T k+ T k ; f <k<n T T n ; f k = () n: Snce I n represents the tme nterval between the last request arrval and the end of the nvestgatng tme span, t s also called as the Watng Tme. P n 2. Average Request Arrval Interval (ARAI): The ARAI s defned as k= Ik =(n ) when n >. ARAI does not exst when n = snce t ndcates only one request arrval wthn tme span T and thus we set t as. For the buffer management, three buffer states and three tmng concepts are defned respectvely as follows.. Constructon State & Start-Tme: When an ntal buffer s allocated upon the arrval of a request, the buffer s flled whle the request s beng served, expectng that the data cached n the buffer could serve closely followed requests for the same object. The sze of the buffer may be adjusted to cache less or more data before t s frozen. Before the buffer sze s frozen, the buffer s n the Constructon State. Thus, the Start-Tme S j of a buffer B j, the j-th buffer allocated for object, s defned as the arrval tme of the last request before the buffer sze s frozen. The requests arrvng n a buffer's Constructon State are called as the resdent requests of ths buffer and the buffer s called as the resdent buffer of these requests. 6

8 Note that f no buffer exsts for a requested object, a frst buffer wth superscrpt j = s allocated. Subsequent buffer allocatons use monotoncally ncreasng js even f the mmedate precedng buffer has been released. Therefore, only after all buffers of the object run to the end, s t possble to reset j and start from agan. 2. Runnng State & Runnng-Dstance: After the buffer freezes ts sze t serves as a runnng wndow of a streamng sesson and moves along wth the streamng sesson. Therefore, the state of the buffer s called the Runnng State. The Runnng-Dstance of a buffer s defned as the dstance n tme between the start-tme of a runnng buffer and the start-tme of ts precedng runnng buffer. We used j to denote the Runnng-Dstance of B j. Note that for the frst buffer allocated to an object, D s equal to the length of object : L, assumng a complete vewng scenaro. Snce we are encouragng sharng among buffers, clents served from B j are also served from any precedng buffers that are stll n runnng state. Ths requres that the runnng-dstance of B j equals to the tme dfference wth the closest precedng buffer n runnng state. Mathematcally, wehave: D j = 8 < : L ; f j = S j Sm ; f j>; (2) where m<jand S m s the start tme of the closest precedng buffer n runnng state. 3. Idle State & End-Tme: When the runnng wndow reaches the end of the streamng sesson, the buffer enters the Idle State, whch s a transent state that allows the buffer to be reclamed. The End-Tme of a buffer s defned as the tme when a buffer enters dle state and s ready to be reclamed. The End-Tme of the buffer B j, denoted as Ej s defned as: E j = 8 < : Sj + L ; f j = mn(s latest + D j ;Sj + L ) f j>: (3) S latest denotes the start tme of the latest runnng buffer for object. Here, E j s dynamcally updated upon the formng of new runnng buffers. The detaled updatng procedure of s descrbed n the followng secton. 3.2 Shared Runnng Buffer (SRB) Algorthm For an ncomng request to the object, the SRB algorthm works as follows: ) If the latest runnng buffer of the object s cachng the prefx of the object, the request s served drectly from all the exstng runnng buffers of the object. 2) Otherwse, (a) If there s enough memory, a new runnng buffer of a predetermned sze T s allocated. The request s served from the new runnng buffer and all exstng runnng buffers of the object. (b) If there s not enough memory, the SRB buffer replacement algorthm (see Secton 3.2.3) s nvoked to ether re-allocate an exstng runnng buffer to the request or serve ths request wthout cachng. 3) Update the End-Tmes of all exstng buffers of the object based on Eq. (3). Durng the process of the SRB algorthm, parts of a runnng buffer could be dynamcally reclamed as descrbed n Secton due to the request termnaton and the buffer s dynamcally managed based on the user access pattern through a lfecycle of three states as descrbed n Secton

9 Meda Poston T (a) Access Tme Meda Poston T (b) Access Tme Meda Poston T T Access Tme (c) Fgure 3: SRB Memory Allocaton: the Intal Buffer Freezes ts Sze (a) (b) (c) 3.2. SRB Buffer Lfecycle Management Intally, a runnng buffer s allocated wth a predetermned sze of T. Startng from the Constructon State, the buffer then adjusts ts sze by gong through a three-state lfecycle management process as descrbed n the followng. ffl Case : the buffer s n the Constructon State. The proxy makes a decson at the end of T as follows. If ARAI =, whch ndcates that there s only one request arrval so far, the ntal buffer enters the Idle State (case 3) mmedately. For ths request, the proxy acts as a bypass server,.e., content s passed to the clent wthout cachng. Ths scheme gves preference to more frequently requested objects n the memory allocaton. Fgure 3(a) llustrates ths stuaton. The shadow ndcates the allocated ntal buffer, whch s reclamed at T. If I n >ARAI(I n s the watng tme), the ntal buffer s shrunk to the extent thatthe most recent request can be served from the buffer. Subsequently, the buffer enters the Runnng State (case 2). Ths runnng buffer then serves as a shftng wndow and run to the end. Fgure 3(b) llustrates an example. Part of the ntal buffer s reclamed at the end of T. Ths scheme performs well for perodcally arrved request groups. If I n» ARAI, the ntal buffer mantans the constructon state and contnues to grow to the length of T, where T = T I n + ARAI, expectng that a new request arrves very soon. At T,theARAI and I n are recalculated and the above procedure repeats. Eventually, when the request to the object becomes less frequent, the buffer wll freeze ts sze and enter the Runnng State (case 2). In the extreme case, the full length of the meda object s cached n the buffer. In ths case, the buffer also freezes and enters the runnng state (a statc runnng). For most frequently accessed objects, ths scheme ensures that the requests to these objects are served from the proxy drectly. Fgure 3(c) llustrates ths stuaton. The ntal buffer has been extended beyond the sze of T for the frst tme. The buffer expanson s bounded by the avalable memory n the proxy. When the avalable memory s exhausted, the buffer freezes ts sze and enters the runnng state regardless of future request arrvals. 8

10 ffl Case 2: the buffer s n the Runnng State. After a buffer enters the runnng state, t starts runnng away from the begnnng of the meda object and subsequent requests can not be served completely from the runnng buffer. In ths case, a new buffer of an ntal sze T s allocated and goes through ts own lfecycle startng from case. Subsequent requests are served from the new buffer as well as ts precedng runnng buffers. Fgure 4 llustrates the maxmal data sharng n the SRB algorthm. The requests R k+ to R n are served smultaneously from B and B 2. Late requests are served from all exstng runnng buffers. Note that except for the frst buffer, the other buffers do not run to the end of the object. R Meda Poston B 2 R 2... R k R k+ R n B Fgure 4: Multple Runnng Buffers Access Tme The Runnng-Dstance and End-Tme are determned based on Eq. (2) and Eq. (3) respectvely for any buffer enterng the Runnng State. In addton, the End-Tmes of ts precedng runnng buffers need to be modfed accordng to the arrval tme of the latest request, as shown n Eq. (3). Fgure shows an example scenaro for the updatng of End-Tmes. Note that E 2 s updated when B 3 s started, at whch tme the runnng dstance of B 2 s extended (dashed lne). When a buffer runs to ts End-Tme, tenters the Idle State (case 3). ffl Case 3: the buffer s n the Idle State. When a buffer enters the Idle State, t s ready for reclamaton. In the above algorthm, the tme span T (whch s the ntal buffer sze) s determned based on the object length. Typcally, a Scale factor (e.g., /2 to /32) of the orgn length s used. To prevent a extremely large or small buffer sze, the buffer sze s bounded by a upper bound Hgh-Bound and a lower bound Low-Bound. These bounds are dependent on the streamng rate to allow the ntal buffer to cache a reasonable porton (e.g., mnute to mnutes) of meda objects. The algorthm requres the clent be able to lsten to multple channels at the same tme: once a request s posted, t should be able to receve data from all the ongong runnng buffers of that object smultaneously. 9

11 Meda poston B B 2 B 3 Access tme S S 2 S 3 E 3 =S 3 +D 3 D 3 E 2 =S 2 +D 2 E 2 =S 3 +D 2 D 2 Fgure : Example on Updatng of End-Tme SRB Dynamc Reclamaton Memory reclamaton of a runnng buffer s trggered by two dfferent types of sesson termnatons: complete sesson termnaton and premature sesson termnaton. In complete sesson termnaton, a sesson termnates only when the delvery of the whole meda object s completed. Assumng that R s the frst request beng served by a runnng buffer, when R reaches the end of the meda object, the resdent buffer of R s reclamed as follows. ffl If the resdent buffer s the only buffer runnng for the meda object, the resdent buffer enters the Idle State. In ths state, the buffer mantans ts content untl all the resdent requests reach the end of the sesson, at whch tme the buffer s released. ffl If the resdent buffer s not the only buffer runnng, that s, there are succeedng runnng buffers, the buffer enters the Idle State and mantans ts content untl ts End-Tme. Note that the End-Tme may have be updated by succeedng runnng buffers. Premature sesson termnaton s much more complcated. In ths case, a request that arrves later may termnate earler. Consder a group of consecutve requests R to R n, the sesson for one of the requests, say R j, where <j<n, termnates before everyone else. The stuaton s handled as follows. ffl If R j s served from the mddle of ts resdent buffer, that s, there are precedng and succeedng requests served from the same runnng buffer, the resdent buffer mantan ts current state and R j s deleted from all ts assocated runnng buffers. Fgure 6(a) and (a') show the buffer stuaton before and after R j s termnated, respectvely. ffl If R j s served from the head of ts resdent buffer as shown n Fgure 6(b), the request s deleted from all of ts assocated runnng buffers. The resdent buffer enters the Idle State for

12 a tme perod of I. Durng ths tme perod, the content wthn the buffer s moved from R j+ to the head. At the end of the tme perod I, the buffer space from the tal to the last served request s released and the buffer enters the Runnng State agan as shown n Fgure 6(b'). ffl If R j s served at the tal of a runnng buffer, two scenaros are further consdered. After deletng the R j from the request lst of ts resdent buffer, f the request lst s not empty, then do nothng. Alternatvely, the algorthm can choose to shrnk the buffer to the extent that R j s served from the buffer assumng R j s a resdent request of the same buffer. In ths case, the End-Tme of the succeedng runnng buffers needs to be adjusted. If R j s at the tal of the last runnng buffer as shown n Fgure 6(c), the buffer s shrunk to the extent that R j s the last request served from the buffer. R j s deleted from the request lst. Subsequently, the buffers run as shown n Fgure 6(c'). Stream of object Stream of object k B k B j R (a) (a ) Stream of object j R (b) k B Stream of object k B (b )... Stream of object Stream of object k B k B j R (c) (c ) Fgure 6: SRB Memory Reclamaton: Dfferent Stuatons of Sesson Termnaton SRB Buffer Replacement Polces The replacement polcy s mportant n the sense that the avalable memory s stll scarce compared to the sze of vdeo objects. So to effcently use the lmted resources s crtcal to acheve the best performance gan. In ths secton, we propose popularty-based replacement polces for the SRB meda cachng algorthm. The basc dea s descrbed as follows:

13 ffl When a request arrves whle there s no avalable memory, all the objects that have on-gong streams n memory are ordered accordng to ther populartes calculated n a certan past tme perod. If the object beng demanded has a hgher popularty than the least popular object n memory, then the latest runnng buffer of the least popular object s released, and the space s re-allocated to the new request. Those requests wthout runnng buffers do not buffer ther data at all. In ths case, theoretcally, they are assumed to have no memory consumpton. Alternatvely, the system can choose to start a memoryless sesson n whch the proxy bypasses the content to the clent wthout cachng. Ths s called a non-replacement polcy. We have evaluated the performances of both two polces by smulatons n the later secton. It s shown that a straghtforward non-replacement polce may acheve smlar performance gven long enough system runnng tme. 3.3 (PSRB) Meda Delverng Algorthm Snce the proxy has fnte amount of memory space, t s possble that the proxy serves as a bypass server wthout cachng concurrent sessons. The SRB algorthm prohbts the sharng of such sessons, whch may lead to excessve server access f there are ntensve request arrvals to many dfferent objects. To solve ths problem, the SRB algorthm can be extended to a patchng SRB (PSRB) algorthm whch enables the sharng of such bypass sessons. Ths s related to the sesson sharng algorthms as dscussed n Secton 2.2. It s mportant to note that PSRB scheme makes the memory-based SRB algorthm work wth the memoryless patchng algorthm. Meda Poston B 2 b a R R 2 R 3 R 4 R R 6 R 7 R 8 B Fgure 7: PSRB Cachng Example Access Tme Fgure 7 llustrates a PSRB scenaro. The frst runnng buffer B has been formed for requests R to R. No buffer s runnng for R6 snce t does not have a close neghborng request. However, a patchng sesson has been started to retreve the absent prefx n B from the content server. At ths tme, request R 6 s served from both the patchng sesson and B untl the mssng prefx s patched. Then, R 6 s served from B only (the sold lne for R 6 stops n the fgure). When R 7 and R 8 arrve and form the second runnng buffer B 2, they are served from B and B 2 2

14 as descrbed n the SRB algorthm. In addton, they are also served from the patchng sesson ntated for R 6. Note that the patchng sesson for R6 s transent, or we can thnk of t as a runnng buffer sesson wth zero buffer sze. As evdent from Fgure 7, the fllng of B 2 does not cause server traffc between poston a and b (no sold lne between a and b) snceb 2 s flled from the patchng sesson for R 6. Thus, sharng the patchng sesson further reduces the the number of server accesses for R 7 and R8. By usng more clent-sde storage, PSRB tres to maxmze the data sharng among concurrent sessons n order to mnmze the server-to-proxy traffc. 4 Workload Analyss To evaluate the performance of the proposed algorthms and to compare them wth pror solutons, we conduct smulatons based on several workloads. Both synthetc workloads and a real workload extracted from enterprse meda server logs are used. We desgn three synthetc workloads. The frst smulates accesses to meda objects n the Web envronment nwhch the length of the vdeo vares from short ones to longer ones. The second smulates the vdeo access n a vdeo-on-demand (VOD) envronment n whch only longer streams are served. Both workloads assume complete vewng clent sessons. We use WEB and P VOD as the name of these workloads. These workloads assume N a Zpf-lke dstrbuton (p = f = = f ;f == ff ) for the popularty of the meda objects. They also assume request nter arrval to follow theposson dstrbuton (p(x; ) =e Λ ( ) x =(x!);x = ; ; 2:::). In the case of vdeo accessng n the Web envronment, clents accesses to vdeos may be ncomplete, that s, a sesson may termnate before the whole meda object s delvered. We smulate ths scenaro by desgnng a partal vewng workload based on the WEB workload. In ths workload, called PARTIAL, 8% of the sessons termnate before 2% of the accessed objects s delvered. For the real workload, we obtan logs from HP Corporate Meda Solutons, coverng the perod from Aprl through Aprl, 2. Durng these ten days, there were two servers runnng Wndows Meda Server TM, servng contents to clents around the world wthn HP ntranet. The contents nclude vdeos, wth the coverage of keynote speeches at varous corporate and ndustry events, messages from the company's management, product announcements, tranng vdeos and product development courses for employees, etc. Ths workload s called REAL. A detaled analyss of the overall characterstcs of the logs from the same servers coverng dfferent tme perods can be found n [6]. 4. Workload Characterstcs Table lsts some statstcs of the four workloads. For the WEB workload, there s a total of 4 unque meda objects, wth a total sze of GB, stored on the server. The length of the meda objects ranges from 2 mnutes to 2 hours. The request nter-arrval follows a Posson dstrbuton wth = 4 seconds and ff = :47 (accordng to[7]) of Zpf-lke dstrbuton for object populartes. The meda objects are coded and streamed at 26 Kbps. The total number of requests s 88. The smulaton lasts 874 seconds or roughly 24 hours. The Low-Bound and Hgh-Bound for the ntal buffer sze are set as 2 MB and 6 MB respectvely. For the PARTIAL workload, 8% of the requests vew only 2% of the requested streamng meda objects and then prematurely termnate. Both the partal-vewng requests and the partal-vewed objects are randomly dstrbuted. Other parameters of the syntheszed trace are dentcal to those 3

15 Workload Number of Number of Total Object Posson Zpf Smulaton name requests objects sze (GB) length (mn) ff duraton (day) WEB ο PARTIAL ο VOD ο REAL ο3 - - Table : Workload Statstcs of WEB as shown n Table. For the VOD workload, the number of unque objects stored on the server s, whch accounts to total sze of 49 GB. The length of the objects ranges from hour to 2 hours. The request nterarrval follows a Posson dstrbuton wth = 6 seconds. The Zpf-lke dstrbuton for object popularty s set as ff = :73 (accordng to []). The meda objects are coded and streamed at 2 Mbps. The total number of requests s 73. The smulaton runs for 6439 seconds or roughly aweek. The Low-Bound and Hgh-Bound for ntal buffer allocaton are set as 6 MB and 28 MB respectvely. For the REAL workload, there s a total of 43 objects wth a total sze of 2 GB. There are 9 request arrvals n a tme span of seconds or roughly days. Fgure 8(a) shows the dstrbuton of the absolute length of vewng tme (n mnutes). It shows that 83.8% of the sessons last less than mnutes. Fgure 8(b) shows the cumulatve dstrbuton of the percentage of vewng tme wth respect to the full object length. It shows that 6.24% of the requests access less than % of the meda object. Only 9.74% of the requests access the full objects. 8 9 Percentage (%) 6 4 Cumulatve dstrbuton (%) ~ 2~3 4~ 6~8 8~ ~2 2+ Absolute length of vewng tme (mn.) ~ 2~3 4~ 6~7 8~9 Vewng percentage (%) Fgure 8: Real workload (left to rght), (a) Dstrbuton of absolute vewng tme (b) dstrbuton of vewng percentage. 4

16 4.2 Potentally Cachable Content To nd out the amount of potentally cachable content, we evaluate the number of overlapped sessons at each tme nstance durng the smulaton. Spec cally, at one pont n tme, f a sesson s streamng a meda object smlar to at least another sesson, t s called a shared sesson. Note that shared sessons may be streamng d erent portons of the meda object. If a sesson s streamng a meda object that no other sesson s streamng, t s called a not-shared sesson. Fgures 9 and show the accumulated number of sessons n two categores durng full or part of the smulatons for the four workloads. 8 2 Shared Not-Shared Shared Not-Shared Sessons Sessons Tme (s) Tme (s) Fgure 9: Potentally Cachable Content: WEB (left) and VOD (rght) workloads 3 6 Shared Not-Shared Sessons Sessons Shared Not-Shared Tme (s) Tme (s) Fgure : Potentally Cachable Content: PARTIAL (left) and REAL (rght) workloads At the rst sght of these gures, t seems that the potental cachable contents of d erent workloads

17 n the decreasng order should be WEB, VOD, PARTIAL and REAL. But be aware that only WEB and PARTIAL show ther whole duratons n the fgures, whle VOD and REAL only show a part of ther full duratons snce ther duratons are too large to be shown on these fgures. Recall that the duraton of PARTIAL s 7 days, whle the duraton of REAL s days. Combnng these factors, we expect that the WEB workload presents the greatest amount of potentally cachable contents, followed by VOD, REAL and PARTIAL workloads n that order. 4.3 Shared Sesson Durng the course of a streamng sesson of an object, when there are other requests accessng the same object, ths porton of the streamng sesson s shared. The actual cachng beneft depends on the number of sharng requests. For example, the beneft of cachng an on-gong sessons smultaneously shared by two requests s dfferent from that by three requests. To further evaluate the beneft the proxy system can get by cachng for shared sessons, we obtan the dstrbuton of the number of requests on the shared sessons. Ths statstcs s collected as follows. At each tme nstance (second), we recode the numbers of on-gong sessons that are shared by dfferent numbers of sharng requests. These numbers are accumulated n bns representng numbers of sharng requests. At the end of the smulaton, the number n each bn s dvded by thetotal smulaton duraton n second, thus obtanng a hstogram of the averaged number of shared sessons at any tme nstance. Fgure (a) shows the dstrbuton n full range on the number of sharng requests, and (b) shows the range from 2 to 2 n more detal. For a gven pont (x; y) on the curves, y ndcates the average number of on-gong sessons that are shared by x number of requests. 3 2 WEB VOD PARTIAL REAL 3 2 WEB VOD PARTIAL REAL Average Number of Shared Sessons 2 Average Number of Shared Sessons Number of Sharng Requests Number of Sharng Requests Fgure : Shared request hstogram. (a) full, (b) part. We fnd that the number of sharng sessons ranges predomnately from 2 to 2. Snce the sessons n the VOD workload last the longest, t s expected that ts average number of shared sessons s the largest among the group of workloads. PARTIAL s the partal vewng case of WEB, thus the number of shared sessons n PARTIAL should be less than that n WEB workload. 6

18 Performance Evaluaton. Evaluaton Metrcs We have mplemented an event-drven smulator to model a proxy's memory cachng behavors. Snce object ht rato or ht rato s not sutable for evaluatng the cachng performance of the streamng meda, we use server traffc reducton rate (shown as "bandwdth reducton" n the fgures) to evaluate the performance of the proposed cachng algorthms. If the algorthms are employed on a server, ths parameter ndcates the dsk I/O traffc reducton rate. Usng SRB or PSRB algorthms, a clent needs to lsten to multple channels for maxmal sharng of cached data n the proxy's memory. We measure the traffc between the proxy and the clent n terms of the average clent channel requrement. Ths s an averaged number of channels the clents are lstenng to durng the sessons. Snce the clents are lstenng to earler on-gong sessons, storage s necessary at the clent sde to buffer the data before ts presentaton. We use the average clent storage requrement n percentage of the full sze of the meda object to ndcate the storage requrement on the clent sde. The effectveness of the algorthms s studed by smulatng dfferent scale factors for the allocaton of the ntal buffer sze and varyng memory cache capactes. The streamng rate s assumed to be constant for smplcty. The smulatons are conducted on HP workstaton x4, wth GHz CPU and GB memory. For each smulaton, we compare a set of seven algorthms n three groups. The frst group contans the bufferng schemes whch nclude dynamc bufferng and nterval cachng. The second group contans the patchng algorthms, namely greedy patchng, grace patchng and optmal patchng algorthms. The thrd group contans the two shared runnng buffer algorthms proposed n ths paper..2 Performance on Synthetc Workloads We consder complete vewng scenaro for streamng meda cachng n both Web envronments and VOD envronments. We also smulate the partal vewng scenaro for web envronment. There are no partal vewng cases for meda delvery n the VOD envronment..2. Complete Vewng Workload of Web Meda Objects Frst, we evaluate the cachng performance wth respect to ntal buffer sze. Wth a fxed memory capacty of GB, the ntal buffer sze vares from to /32 of the length of the meda object. For each scale factor, an ntal buffer of dfferent sze s allocated f the meda object length s dfferent. The server traffc reducton, the average clent channel requrement and the average clent storage requrement are recorded n the smulaton. The results are plotted n Fgure 2. Fgure 2(a) shows the server traffc reducton acheved by each algorthm. Note that PSRB acheves the best reducton and SRB acheves the next best reducton except optmal patchng. RB cachng acheves the least amount of reducton. As expected, the performance of the three patchng algorthms does not depend on the scale factor for allocatng of the ntal buffer. Nether does that of nterval cachng snce t allocates buffers based on access ntervals. 7

19 7 3 Bandwdth Reducton (%) Average Clent Channel Requrement Average Clent Storage Requrement (%) /Scale /Scale /Scale Fgure 2: WEB workload (left to rght): (a) Server Traffc Reducton, (b) Average Clent Channel and (c) Storage Requrement(%) wth GB Proxy Memory For the runnng buffer schemes, we notce some varaton n the performance wth respect to the scale factor. In general, the varatons are lmted. To a certan extent, the performance gan of the bandwdth reducton s a trade-off between the number of runnng buffers and the sze of runnng buffers. A larger buffer ndcates that more requests can be served from the t. However, a larger buffer ndcates less memory space left for other requests. Ths n turn leads to more server accesses snce there s no avalable proxy memory. On the other hand, a smaller buffer may serve a smaller number of requests but t leaves more proxy memory space to allocate for other requests. Fgure 2 (b) and (c) show the average channel and storage requrement on the clent. Note that optmal patchng acheves the better server traffc reducton by payng the penalty of mposng the bggest number of clent channels requred. Comparatvely, PSRB and SRB requres 3ο6% less clent channels whle achevng smlar or better server traffc reducton. PSRB allows sesson sharng even when memory space s not avalable. It s therefore expected that PSRB acheves the maxmal server traffc reducton. In the mean tme, t also requres the maxmum clent sde storage and clent channels. On the other hand, SRB acheves 6 percentage pont less traffc reducton than PSRB, but the requrement on clent channel and storage s sgnfcantly lower Bandwdth Reducton (%) Average Clent Channel Requrement Average Clent Storage Requrement (%) Memory Cache Sze (Gbytes) Memory Cache Sze (Gbytes) Memory Cache Sze (Gbytes) Fgure 3: WEB workload (left to rght): (a)bandwdth Reducton, (b)average Clent Channel and (c)storage Requrement(%) wth the scale of /4 8

20 We now nvestgate the performance of dfferent algorthms wth respect to varous memory capactes on the proxy. In ths smulaton, we use a fxed scale factor of /4 for the ntal buffer sze. Fgure 3 (a) ndcates flat traffc reducton rate for the three patchng algorthms. Ths s expected snce no proxy memory resource s utlzed n patchng. On the other hand, all the other algorthms nvestgated acheve hgher traffc reducton when memory capacty ncreases. It s mportant to note that the proposed two SRB algorthms acheve better traffc reducton than the nterval cachng and runnng buffer schemes. In Fgure 3(b), the clent channel requrement decreases for the PSRB algorthm when the cache capacty ncreases. Ths s agan expected snce more clents are severed from the proxy buffers nstead of proxy patchng sessons. When the cache capacty reaches 4 GB, PSRB requres only 3% of the clent channel needed for the optmal patchng scheme. PSRB also requres less clent storage at ths cache capacty as ndcated n Fgure 3(c). And yet, PSRB acheves more than percentage ponts of traffc reducton comparng to the optmal patchng scheme. For the SRB algorthm, t generally acheves the second best traffc reducton wth even less penalty on clent channel and storage requrements..2.2 Complete Vewng Workloads of VOD Objects To further evaluate the performance of the proposed algorthm n a vdeo delvery envronment wth relatvely longer streamng sessons, we use the VOD workload for the smulaton. Compared wth the result we obtaned n the WEB workload smulaton, the smulaton on VOD workload demonstrates smlar trade-offs wth respect to varyng scale factors. Ths s evdent as shown n Fgure Bandwdth Reducton (%) Average Clent Channel Requrement Average Clent Storage Requrement (%) /Scale /Scale /Scale Fgure 4: VOD workload (left to rght): (a) Bandwdth Reducton, (b) Average Clent Channel and (c) Storage Requrement(%) wth GB Proxy Memory Wth a fxed scale factor for the ntal buffer sze, Fgure (a) llustrates smlar performance gan as acheved for the WEB workload. In addton, smlar penaltes are mposed on each algorthm as s evdent from Fgure (b) and (c). Note that the performance gan for the VOD workload s larger than that of the WEB workload. Ths s manly due to the longer streamng sesson on average n the VOD workload. A longer streamng sesson provdes more opportunty for multple clents to share the sesson. 9

21 4 3 2 Bandwdth Reducton (%) Average Clent Channel Requrement Average Clent Storage Requrement (%) Memory Cache Sze (Gbytes) Memory Cache Sze (Gbytes) Memory Cache Sze (Gbytes) Fgure : VOD workload (left to rght): (a) Bandwdth Reducton, (b) Average Clent Channel and (c) Storage Requrement(%) wth the Scale of /4.2.3 Partal Vewng of Web Workload In streamng meda delvery over the Web, t s possble that some clents termnate the sesson after watchng for a whle from the begnnng of the meda object. It s mportant to evaluate the performance of the proposed algorthm under ths stuaton. Usng the PARTIAL workload as defned n Secton 4, we perform the same smulatons and evaluate the same set of parameters. Fgure 6(a) shows smlar characterstcs as that n Fgure 2. PSRB and SRB stll acheves better traffc reducton. The concluson holds that PSRB uses 6% of the clent channel to acheve percentage pont better traffc reducton compared wth the optmal patchng. 4 7 Bandwdth Reducton (%) Average Clent Channel Requrement /Scale /Scale Fgure 6: PARTIAL workload (left to rght): (a) Bandwdth Reducton and (b) Average Clent Channel Requrement wth GB Proxy Memory In the event that a sesson termnates before t reaches the end of the requested meda object, t s possble that the clent has already downloaded future part of the meda stream whch s no longer needed. To characterze ths wasted delvery from the proxy to the clent, we record average clent waste. It s the percentage of wasted bytes versus the total prefetched data. Fgure 7 shows the 2

22 clent waste statstc. Note that for PARTIAL and REAL workloads, snce both contan premature sesson termnatons, the prefetched data whch s not used n the presentaton are not counted as bytes ht n the calculaton of the server traffc reducton. 2 Average Clent Storage Requrement (%) 2 Average Clent Waste (%) /Scale /SCale Fgure 7: PARTIAL workload (left to rght): (a) Average Clent Storage Requrement(%) and (b) Clent Waste(%) wth GB Proxy Memory As shown n Fgure 7(b), PSRB and SRB have about 42% and % of prefetched data wasted comparng wth % for nterval cachng. Snce the wasted bytes are not counted as ht, ths leads to the lowered traffc reducton rate for PSRB and SRB comparng to that of nterval cache. From another perspectve, nterval cachng does not promote sharng among buffers, hence the clent lstens to one channel only and there s no bufferng of future data. Thus, there s no waste n proxy-to-clent delvery n the event of premature sesson termnaton. We now agan start nvestgaton of the cachng performance wth a fxed scale factor for the ntal buffer sze n Fgure 8. Comparng wth Fgure 2(a), the dstances between the traffc reducton curves between PSRB, SRB and nterval cachng become much smaller n general. Ths renforces the observaton above that PSRB and SRB may lead to more wasted bytes n the partal vewng cases. In addton, the grace patchng achevng almost no traffc reducton shows ts ncapablty n dealng wth the partal vewng stuaton. The reason mght be that the new sesson started by the grace patchng, whch s supposed to be a complete sesson, often termnates when 2% of the meda object s delvered. Snce the duraton of the sesson s short, t s less lkely that a new request to the same meda object s receved. In Fgure 8(b), PSRB demonstrates monotonc decreasng of average clent channel requrement when memory capacty ncreases. Ths s due to the fact that there s a fewer number of zero-szed runnng buffers wth ncreasng proxy memory capacty. Smlarly, as shown n Fgure 9, the clent storage requrement and average clent waste also decrease snce a fewer number of patchng s requred. 2

SRB: Shared Running Buffers in Proxy to Exploit Memory Locality of Multiple Streaming Media Sessions

SRB: Shared Running Buffers in Proxy to Exploit Memory Locality of Multiple Streaming Media Sessions SRB: Shared Runnng Buffers n Proxy to Explot Memory Localty of Multple Streamng Meda Sessons Songqng Chen,BoShen, Yong Yan, Sujoy Basu, and Xaodong Zhang Department of Computer Scence Moble and Meda System

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

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

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

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

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

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

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

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

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

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

ARTICLE IN PRESS. Signal Processing: Image Communication

ARTICLE IN PRESS. Signal Processing: Image Communication Sgnal Processng: Image Communcaton 23 (2008) 754 768 Contents lsts avalable at ScenceDrect Sgnal Processng: Image Communcaton journal homepage: www.elsever.com/locate/mage Dstrbuted meda rate allocaton

More information

Cache Performance 3/28/17. Agenda. Cache Abstraction and Metrics. Direct-Mapped Cache: Placement and Access

Cache Performance 3/28/17. Agenda. Cache Abstraction and Metrics. Direct-Mapped Cache: Placement and Access Agenda Cache Performance Samra Khan March 28, 217 Revew from last lecture Cache access Assocatvty Replacement Cache Performance Cache Abstracton and Metrcs Address Tag Store (s the address n the cache?

More information

Optimized caching in systems with heterogeneous client populations

Optimized caching in systems with heterogeneous client populations Performance Evaluaton 42 (2000) 163 185 Optmzed cachng n systems wth heterogeneous clent populatons Derek L. Eager a,, Mchael C. Ferrs b, Mary K. Vernon b a Department of Computer Scence, Unversty of Saskatchewan,

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

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

Run-Time Operator State Spilling for Memory Intensive Long-Running Queries

Run-Time Operator State Spilling for Memory Intensive Long-Running Queries Run-Tme Operator State Spllng for Memory Intensve Long-Runnng Queres Bn Lu, Yal Zhu, and lke A. Rundenstener epartment of Computer Scence, Worcester Polytechnc Insttute Worcester, Massachusetts, USA {bnlu,

More information

A Fast Visual Tracking Algorithm Based on Circle Pixels Matching

A Fast Visual Tracking Algorithm Based on Circle Pixels Matching A Fast Vsual Trackng Algorthm Based on Crcle Pxels Matchng Zhqang Hou hou_zhq@sohu.com Chongzhao Han czhan@mal.xjtu.edu.cn Ln Zheng Abstract: A fast vsual trackng algorthm based on crcle pxels matchng

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

X- Chart Using ANOM Approach

X- Chart Using ANOM Approach ISSN 1684-8403 Journal of Statstcs Volume 17, 010, pp. 3-3 Abstract X- Chart Usng ANOM Approach Gullapall Chakravarth 1 and Chaluvad Venkateswara Rao Control lmts for ndvdual measurements (X) chart are

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

EECS 730 Introduction to Bioinformatics Sequence Alignment. Luke Huan Electrical Engineering and Computer Science

EECS 730 Introduction to Bioinformatics Sequence Alignment. Luke Huan Electrical Engineering and Computer Science EECS 730 Introducton to Bonformatcs Sequence Algnment Luke Huan Electrcal Engneerng and Computer Scence http://people.eecs.ku.edu/~huan/ HMM Π s a set of states Transton Probabltes a kl Pr( l 1 k Probablty

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

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

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

Improvement of Spatial Resolution Using BlockMatching Based Motion Estimation and Frame. Integration

Improvement of Spatial Resolution Using BlockMatching Based Motion Estimation and Frame. Integration Improvement of Spatal Resoluton Usng BlockMatchng Based Moton Estmaton and Frame Integraton Danya Suga and Takayuk Hamamoto Graduate School of Engneerng, Tokyo Unversty of Scence, 6-3-1, Nuku, Katsuska-ku,

More information

USING GRAPHING SKILLS

USING GRAPHING SKILLS Name: BOLOGY: Date: _ Class: USNG GRAPHNG SKLLS NTRODUCTON: Recorded data can be plotted on a graph. A graph s a pctoral representaton of nformaton recorded n a data table. t s used to show a relatonshp

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

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

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

Content Classification for Caching under CDNs

Content Classification for Caching under CDNs Content Classfcaton for Cachng under CDNs George Palls 1,2, Charlaos homos 2, Konstantnos Stamos 2, Athena Vakal 2, George Andreads 3 1 Department of Computer Scence, Unversty of Cyprus 2 Department of

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

3D vector computer graphics

3D vector computer graphics 3D vector computer graphcs Paolo Varagnolo: freelance engneer Padova Aprl 2016 Prvate Practce ----------------------------------- 1. Introducton Vector 3D model representaton n computer graphcs requres

More information

Analytic Evaluation of Quality of Service for On-Demand Data Delivery

Analytic Evaluation of Quality of Service for On-Demand Data Delivery Analytc Evaluaton of Qualty of Servce for On-Demand Data Delvery Hongfe Guo Haonan Tan ( guo@cs.wsc.edu) (haonan@cs.wsc.edu) Abstract Qualty of servce (QoS) measured as balkng probablty and average watng

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

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

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

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

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

Reducing I/O Demand in Video-On-Demand Storage Servers. from being transmitted directly from tertiary devices.

Reducing I/O Demand in Video-On-Demand Storage Servers. from being transmitted directly from tertiary devices. Reducng I/O Demand n Vdeo-On-Demand Storage Servers Leana Golubchk y John C.S. Lu z Rchard Muntz x Abstract Recent technologcal advances have made multmeda on-demand servces, such as home entertanment

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

User Authentication Based On Behavioral Mouse Dynamics Biometrics

User Authentication Based On Behavioral Mouse Dynamics Biometrics User Authentcaton Based On Behavoral Mouse Dynamcs Bometrcs Chee-Hyung Yoon Danel Donghyun Km Department of Computer Scence Department of Computer Scence Stanford Unversty Stanford Unversty Stanford, CA

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

Scalable Download Protocols

Scalable Download Protocols Scalable Download Protocols A Thess Submtted to the College of Graduate Studes and Research In Partal Fulfllment of the Requrements For the Degree of Doctor of Phlosophy In the Department of Computer Scence

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

Some Advanced SPC Tools 1. Cumulative Sum Control (Cusum) Chart For the data shown in Table 9-1, the x chart can be generated.

Some Advanced SPC Tools 1. Cumulative Sum Control (Cusum) Chart For the data shown in Table 9-1, the x chart can be generated. Some Advanced SP Tools 1. umulatve Sum ontrol (usum) hart For the data shown n Table 9-1, the x chart can be generated. However, the shft taken place at sample #21 s not apparent. 92 For ths set samples,

More information

Fast Retransmission of Real-Time Traffic in HIPERLAN/2 Systems

Fast Retransmission of Real-Time Traffic in HIPERLAN/2 Systems Fast Retransmsson of Real-Tme Traffc n HIPERLAN/ Systems José A Afonso and Joaqum E Neves Department of Industral Electroncs Unversty of Mnho, Campus de Azurém 4800-058 Gumarães, Portugal {joseafonso,

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

Improved Resource Allocation Algorithms for Practical Image Encoding in a Ubiquitous Computing Environment

Improved Resource Allocation Algorithms for Practical Image Encoding in a Ubiquitous Computing Environment JOURNAL OF COMPUTERS, VOL. 4, NO. 9, SEPTEMBER 2009 873 Improved Resource Allocaton Algorthms for Practcal Image Encodng n a Ubqutous Computng Envronment Manxong Dong, Long Zheng, Kaoru Ota, Song Guo School

More information

Reducing Frame Rate for Object Tracking

Reducing Frame Rate for Object Tracking Reducng Frame Rate for Object Trackng Pavel Korshunov 1 and We Tsang Oo 2 1 Natonal Unversty of Sngapore, Sngapore 11977, pavelkor@comp.nus.edu.sg 2 Natonal Unversty of Sngapore, Sngapore 11977, oowt@comp.nus.edu.sg

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

SLAM Summer School 2006 Practical 2: SLAM using Monocular Vision

SLAM Summer School 2006 Practical 2: SLAM using Monocular Vision SLAM Summer School 2006 Practcal 2: SLAM usng Monocular Vson Javer Cvera, Unversty of Zaragoza Andrew J. Davson, Imperal College London J.M.M Montel, Unversty of Zaragoza. josemar@unzar.es, jcvera@unzar.es,

More information

A fair buffer allocation scheme

A fair buffer allocation scheme A far buffer allocaton scheme Juha Henanen and Kalev Klkk Telecom Fnland P.O. Box 228, SF-330 Tampere, Fnland E-mal: juha.henanen@tele.f Abstract An approprate servce for data traffc n ATM networks requres

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

Adaptive Power-Aware Prefetch in Wireless Networks

Adaptive Power-Aware Prefetch in Wireless Networks Adaptve Power-Aware Prefetch n Wreless Networks Langzhong Yn, Student Member, IEEE, and Guohong Cao, Member, IEEE Abstract Most of the prefetch technques used n the current cache management schemes do

More information

Harfoush, Bestavros, and Byers, Robust Identfcaton of Shared Losses Usng End-to-End Uncast Probes 2 Introducton One of the defnng prncples of the netw

Harfoush, Bestavros, and Byers, Robust Identfcaton of Shared Losses Usng End-to-End Uncast Probes 2 Introducton One of the defnng prncples of the netw Robust Identfcaton of Shared Losses Usng End-to-End Uncast Probes Λ Khaled Harfoush Azer Bestavros John Byers harfoush@cs.bu.edu best@cs.bu.edu byers@cs.bu.edu Computer Scence Department Boston Unversty

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

FRES-CAR: An Adaptive Cache Replacement Policy

FRES-CAR: An Adaptive Cache Replacement Policy FRES-CAR: An Adaptve Cache Replacement Polcy George Palls, Athena Vaal, Eythms Sdropoulos Department of Informatcs Arstotle Unversty of Thessalon, 54124, Thessalon, Greece gpalls@ccf.auth.gr, {avaal, eythms}@csd.auth.gr

More information

Network-Driven Layered Multicast with IPv6

Network-Driven Layered Multicast with IPv6 Network-Drven Layered Multcast wth IPv6 Ho-pong Sze and Soung C. Lew Department of Informaton Engneerng, The Chnese Unversty of Hong Kong, Shatn, N.T., Hong Kong {hpsze8, soung}@e.cuhk.edu.hk Abstract.

More information

FORESIGHTED JOINT RESOURCE RECIPROCATION AND SCHEDULING STRATEGIES FOR REAL-TIME VIDEO STREAMING OVER PEER-TO-PEER NETWORKS

FORESIGHTED JOINT RESOURCE RECIPROCATION AND SCHEDULING STRATEGIES FOR REAL-TIME VIDEO STREAMING OVER PEER-TO-PEER NETWORKS FORESIGHTED JOINT RESOURCE RECIPROCATION AND SCHEDULING STRATEGIES FOR REAL-TIME VIDEO STREAMING OVER PEER-TO-PEER NETWORKS Sunghoon Ivan Lee, Hyunggon Park, and Mhaela van der Schaar Electrcal Engneerng

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

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

On the Exact Analysis of Bluetooth Scheduling Algorithms

On the Exact Analysis of Bluetooth Scheduling Algorithms On the Exact Analyss of Bluetooth Schedulng Algorth Gl Zussman Dept. of Electrcal Engneerng Technon IIT Hafa 3000, Israel glz@tx.technon.ac.l Ur Yechal Dept. of Statstcs and Operatons Research School of

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

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

Adaptive Load Shedding for Windowed Stream Joins

Adaptive Load Shedding for Windowed Stream Joins Adaptve Load Sheddng for Wndowed Stream Jons Bu gra Gedk College of Computng, GaTech bgedk@cc.gatech.edu Kun-Lung Wu, Phlp Yu T.J. Watson Research, IBM {klwu,psyu}@us.bm.com Lng Lu College of Computng,

More information

NUMERICAL SOLVING OPTIMAL CONTROL PROBLEMS BY THE METHOD OF VARIATIONS

NUMERICAL SOLVING OPTIMAL CONTROL PROBLEMS BY THE METHOD OF VARIATIONS ARPN Journal of Engneerng and Appled Scences 006-017 Asan Research Publshng Network (ARPN). All rghts reserved. NUMERICAL SOLVING OPTIMAL CONTROL PROBLEMS BY THE METHOD OF VARIATIONS Igor Grgoryev, Svetlana

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

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

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

Adaptive Load Shedding for Windowed Stream Joins

Adaptive Load Shedding for Windowed Stream Joins Adaptve Load Sheddng for Wndowed Stream Jons Buğra Gedk, Kun-Lung Wu, Phlp S. Yu, Lng Lu College of Computng, Georga Tech Atlanta GA 333 {bgedk,lnglu}@cc.gatech.edu IBM T. J. Watson Research Center Yorktown

More information

VISUAL SELECTION OF SURFACE FEATURES DURING THEIR GEOMETRIC SIMULATION WITH THE HELP OF COMPUTER TECHNOLOGIES

VISUAL SELECTION OF SURFACE FEATURES DURING THEIR GEOMETRIC SIMULATION WITH THE HELP OF COMPUTER TECHNOLOGIES UbCC 2011, Volume 6, 5002981-x manuscrpts OPEN ACCES UbCC Journal ISSN 1992-8424 www.ubcc.org VISUAL SELECTION OF SURFACE FEATURES DURING THEIR GEOMETRIC SIMULATION WITH THE HELP OF COMPUTER TECHNOLOGIES

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

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

Motivation. EE 457 Unit 4. Throughput vs. Latency. Performance Depends on View Point?! Computer System Performance. An individual user wants to:

Motivation. EE 457 Unit 4. Throughput vs. Latency. Performance Depends on View Point?! Computer System Performance. An individual user wants to: 4.1 4.2 Motvaton EE 457 Unt 4 Computer System Performance An ndvdual user wants to: Mnmze sngle program executon tme A datacenter owner wants to: Maxmze number of Mnmze ( ) http://e-tellgentnternetmarketng.com/webste/frustrated-computer-user-2/

More information

Computer models of motion: Iterative calculations

Computer models of motion: Iterative calculations Computer models o moton: Iteratve calculatons OBJECTIVES In ths actvty you wll learn how to: Create 3D box objects Update the poston o an object teratvely (repeatedly) to anmate ts moton Update the momentum

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

Clustered Multimedia NOD : Popularity-Based Article Prefetching and Placement

Clustered Multimedia NOD : Popularity-Based Article Prefetching and Placement Clustered Multmeda NOD : Popularty-Based Artcle Prefetchng and Placement Y.J.Km, T.U.Cho, K.O.Jung, Y.K.Kang, S.H.Park, K-Dong Chung Department of Computer Scence, Pusan Natonal Unversty, Korea Abstract

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

SAO: A Stream Index for Answering Linear Optimization Queries

SAO: A Stream Index for Answering Linear Optimization Queries SAO: A Stream Index for Answerng near Optmzaton Queres Gang uo Kun-ung Wu Phlp S. Yu IBM T.J. Watson Research Center {luog, klwu, psyu}@us.bm.com Abstract near optmzaton queres retreve the top-k tuples

More information

MobileGrid: Capacity-aware Topology Control in Mobile Ad Hoc Networks

MobileGrid: Capacity-aware Topology Control in Mobile Ad Hoc Networks MobleGrd: Capacty-aware Topology Control n Moble Ad Hoc Networks Jle Lu, Baochun L Department of Electrcal and Computer Engneerng Unversty of Toronto {jenne,bl}@eecg.toronto.edu Abstract Snce wreless moble

More information

Efficient Content Distribution in Wireless P2P Networks

Efficient Content Distribution in Wireless P2P Networks Effcent Content Dstrbuton n Wreless P2P Networs Qong Sun, Vctor O. K. L, and Ka-Cheong Leung Department of Electrcal and Electronc Engneerng The Unversty of Hong Kong Pofulam Road, Hong Kong, Chna {oansun,

More information

An Efficient Garbage Collection for Flash Memory-Based Virtual Memory Systems

An Efficient Garbage Collection for Flash Memory-Based Virtual Memory Systems S. J and D. Shn: An Effcent Garbage Collecton for Flash Memory-Based Vrtual Memory Systems 2355 An Effcent Garbage Collecton for Flash Memory-Based Vrtual Memory Systems Seunggu J and Dongkun Shn, Member,

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

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

Network Coding as a Dynamical System

Network Coding as a Dynamical System Network Codng as a Dynamcal System Narayan B. Mandayam IEEE Dstngushed Lecture (jont work wth Dan Zhang and a Su) Department of Electrcal and Computer Engneerng Rutgers Unversty Outlne. Introducton 2.

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

Goals and Approach Type of Resources Allocation Models Shared Non-shared Not in this Lecture In this Lecture

Goals and Approach Type of Resources Allocation Models Shared Non-shared Not in this Lecture In this Lecture Goals and Approach CS 194: Dstrbuted Systems Resource Allocaton Goal: acheve predcable performances Three steps: 1) Estmate applcaton s resource needs (not n ths lecture) 2) Admsson control 3) Resource

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

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

Efficient QoS Provisioning at the MAC Layer in Heterogeneous Wireless Sensor Networks

Efficient QoS Provisioning at the MAC Layer in Heterogeneous Wireless Sensor Networks Effcent QoS Provsonng at the MAC Layer n Heterogeneous Wreless Sensor Networks M.Soul a,, A.Bouabdallah a, A.E.Kamal b a UMR CNRS 7253 HeuDaSyC, Unversté de Technologe de Compègne, Compègne Cedex F-625,

More information

Efficient Broadcast Disks Program Construction in Asymmetric Communication Environments

Efficient Broadcast Disks Program Construction in Asymmetric Communication Environments Effcent Broadcast Dsks Program Constructon n Asymmetrc Communcaton Envronments Eleftheros Takas, Stefanos Ougaroglou, Petros copoltds Department of Informatcs, Arstotle Unversty of Thessalonk Box 888,

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

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

Support Strong Consistency for Mobile Dynamic Contents Delivery Network

Support Strong Consistency for Mobile Dynamic Contents Delivery Network Nnth IEEE Internatonal Symposum on Multmeda 27 Support Strong Consstency for Moble Dynamc Contents Delvery Networ Zhou Su, Jro Katto and Yasuho Yasuda Graduate School of Scence and Engneerng, Waseda Unversty,

More information

Concurrent Apriori Data Mining Algorithms

Concurrent Apriori Data Mining Algorithms Concurrent Apror Data Mnng Algorthms Vassl Halatchev Department of Electrcal Engneerng and Computer Scence York Unversty, Toronto October 8, 2015 Outlne Why t s mportant Introducton to Assocaton Rule Mnng

More information

Technical Report. i-game: An Implicit GTS Allocation Mechanism in IEEE for Time- Sensitive Wireless Sensor Networks

Technical Report. i-game: An Implicit GTS Allocation Mechanism in IEEE for Time- Sensitive Wireless Sensor Networks www.hurray.sep.pp.pt Techncal Report -GAME: An Implct GTS Allocaton Mechansm n IEEE 802.15.4 for Tme- Senstve Wreless Sensor etworks Ans Koubaa Máro Alves Eduardo Tovar TR-060706 Verson: 1.0 Date: Jul

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

Simulation and Exploration of RCP in the networks

Simulation and Exploration of RCP in the networks EE384Y Smulaton and Exploraton of P n the networks Sprng, 003 Smulaton and Exploraton of P n the networks hanghua He changhua@stanford.edu June 06, 003 Abstract P (ate ontrol Protocol) s a rate-based congeston

More information

Feature Reduction and Selection

Feature Reduction and Selection Feature Reducton and Selecton Dr. Shuang LIANG School of Software Engneerng TongJ Unversty Fall, 2012 Today s Topcs Introducton Problems of Dmensonalty Feature Reducton Statstc methods Prncpal Components

More information