Performance Characterization of a Bluetooth Piconet with Multi Slot Packets aniele Miorani Carlo Caimi Anrea anella To cite this version: aniele Miorani Carlo Caimi Anrea anella. Performance Characterization of a Bluetooth Piconet with Multi Slot Packets. WiOpt 3: Moeling an Optimization in Mobile A Hoc an Wireless Networks Mar 3 Sophia Antipolis France. 6 p. 3. inria-667> HA I: inria-667 https://hal.inria.fr/inria-667 Submitte on 5 Mar HA is a multi-isciplinary open access archive for the eposit an issemination of scientific research ocuments whether they are publishe or not. The ocuments may come from teaching an research institutions in France or abroa or from public or private research centers. archive ouverte pluriisciplinaire HA est estinée au épôt et à la iffusion e ocuments scientifiques e niveau recherche publiés ou non émanant es établissements enseignement et e recherche français ou étrangers es laboratoires publics ou privés.
Performance Characterization of a Bluetooth Piconet with Multi Slot Packets aniele Miorani Carlo Caimi an Anrea anella epartment of Information Engineering University of Paova ia Graenigo 6/B 353 Paova Italy email: aniele.miorani carlo.caimi anrea.zanella @ei.unip.it Abstract In this paper we present a framework for performance evaluation of a Bluetooth piconet using multislot packets. In particular uner some classical assumptions we evelop a moel of a Bluetooth network an erive the complete statistics of the one hop elay an other significant metrics such as the channel utilization parameter. Stability conitions are investigate an it is shown that the use of multi slot packets enlarges the achievable rate region. Simulations results are then shown which valiate the propose analysis. Inex Terms Bluetooth S.A.R. multislot packets performance evaluation I. INTROUCTION Originally born as a wireless replacement for cables connecting electronic evices Bluetooth [] has been gaining a lot of consieration an attention by the scientific community in the last few years [] [3] []. The evelopment of this technology is now focuse on the so-calle Wireless Personal Area Networks (WPANs) which are foreseen as the maor application for Bluetooth evices in the short an mi-term future. While the iffusion of Bluetooth evices is believe to experience a rapi growth in the next years thanks to its expecte low cost (own to US accoring to leaing manufacturers) the success of such a technology will still be linke to its ability to track the eman for avance applications. Inee performance improvement schemes have become an active research fiel in orer to let Bluetooth networks being able to support emaning services like Internet MP3 auio an low-quality vieo. The basic brick to buil up a Bluetooth base network is the so calle piconet a cluster of noes (not more than ) communicating with each other by sharing a common FH channel. Access control to the raio channel is achieve by using a master riven T scheme base on a master slave architecture. Piconets may communicate by sharing on a time ivision basis a evice calle inter piconet unit which acts as gateway among the piconets it belongs to. Even if much attention has been evote to performance evaluation of Bluetooth networks most of the results have been obtaine through numerical simulations. Inee the literature still lacks an in epth analytical investigation of Bluetooth network performance. Not even the analysis of a single piconet with multi slot packets has been complete even if some consierable effort has been evote in the last years [5] [6]. In this paper we aim at proviing a mathematical framework This work was supporte by MIUR within the framework of the PRIMO proect FIRB RBNE8RFY which uner some simplifying but classical assumptions allows us to evaluate the performance obtainable with the use of multi slot packets. In particular we provie the packet elay statistics for one hop transmissions. In this sense our approach resembles that use in [6] from which it iffers for some moel assumptions in particular on the role of the master an on the possibility of having ifferent statistics for the packet length at ifferent noes. Furthermore we provie a characterization of the channel utilization parameter an an in epth investigation of the stability regions achievable uner a given Segmentation An Reassemly (SAR) policy showing that the use of multi slot packets provies an enlargement of the achievable rate region. The paper is organize as follows: Section II eals with the characteristics of the Bluetooth communications profile an escribes the system moel which will be use for the analysis. Section III eals with the performance analysis of a Bluetooth piconet in terms of single hop elay together with an investigation of the achievable capacity regions. Finally Section I presents some concluing remarks an open issues for future work. II. BACKGROUN: THE BUETOOTH TECHNOOGY Bluetooth operates in the GHz ISM unlicense ban proviing a raw bit rate of Mb/s by using a binary Gaussianshape FSK moulation. In orer to reuce interference with other evices operating in the ISM ban Bluetooth aopts a frequency hopping (FH) sprea spectrum technique spanning RF carriers -MHz wie each. In orer to communicate two up to eight Bluetooth units may connect in a small network calle piconet. In each piconet a unit acts as master controlling the channel access by means of a simple polling scheme. Time is ivie into consecutive slots of s each that are use for ownlink (master-to-slave) an uplink (slave-to-master) transmissions alternatively in a time ivision uplex (T) fashion. Namely each time-slot is associate to a hop in the hopping sequence resulting in a nominal hop rate of hop/s. The master can transmit in even numbere time slots whereas o numbere slots are reserve for slaves transmissions. The stanar provies two types of service Asynchronous Connectioness (AC) an Synchronous Connection Oriente (SCO). The latter provies a way of sening elay sensitive services (tipically voice) over Bluetooth an is base on a reservation scheme which estabilishes a virtual circuit between a slave an the master. The other type of link AC provies packet switching on
" % " % g h I O O PSfrag replacements Type M M3 M5 H H3 H5 Fig.. Slot occupancy Multi slot packet transmission Max.payloa length (bytes) $ &' &" "&"&) TABE I PACKET CHARACTERISTICS FOR AC INKS FEC rate!"!"!" the wireless channel; the stanar presents six possible packet types which iffer for both coing scheme applie (either a - + shortene Hamming coe or none) an packet length (-+. an time slots) as epicte in Fig.. A resume of the packet characteristics is reporte in Tab.I. ifferent piconets are associate to inepenent FH channels. This allows more piconets to share the same physical space an spectrum without increasing excessively the mutual interference. Piconet may communicate by sharing a evice on a time ivision basis forming what is commonly referre to as a scatternet. Both a piconet an a scatternet consisting of /. piconets are epicte in Fig.. embee on the chip an thus it results the most attractive choice for low cost an power-aware solutions. Furthermore in the following we will consier AC links only an limit our analysis to the use of unprotecte packets (3576 where 68 9+.+ ). In fact it has been shown [9] that in most operating conitions (namely for not too low signal to noise ratios) the use of 3576 packets results in higher gooput with respect to 35/:6. We start by consiering a piconet consisting of ;= noes ; the network may then be represente by a set of ;?> @; interacting queues. Since the solution for problems of interacting queues is still far from coming we assume that the arrival processes at the various queues are inepenent. To treat also slave to slave communications we will employ the classical tool of statistical routing [] which enables us to get an approximate analysis of the network performance. In other wors the traffic from the master to a particular slave say A will be compute as the sum of the traffic generate in the whole piconet for that particular slave an the various resulting flows will be consiere inepenent. To get a more mathematically comprehensive framework we nee to introuce some preliminary notations. et us enumerate units in a piconet from to ;. et B5C + + ; be the master I. For each +AEC + &+; we use the suffix +A to enote the link between noe an A. Note that in case both an A are slave units the link is virtual since each packet exchange between two slaves has to be route through the master that will forwar the packet to the recipient slave. The arrivals are moele with a Poisson process with bulk arrivals an rate FHGI (expresse in packets/slot). The probabilities of packets generate at noe to noe A being one three an NORQ five slots long are KGI KMGI. an KMGI with I SI T K GUI. We assume that no segmentation an reassembly (SAR) proceures take place when a slave to slave communication passes through the master. A complete characterization of the piconet is thus given by: A. System Moel To carry on our analysis we nee some simplifying assumptions on the consiere network. First of all we limit W BX+Y -+ () our analysis to the simplest polling scheme Pure Roun Robin where (PRR) also referre to as imite- polling. Although many Y is the polynomial en to en traffic matrix whose +A -th entry is efine as: efficient polling schemes have been propose in the last few years [7] [8] at this time available evices implement a Y GI F GI\[ -^X ORQ K GI () basic PRR scheme. This is essentially ue to the necessity of SI T keeping the complexity of the firmware as low as possible in orer to reuce the manufacturing costs an lower the The en to en traffic matrix Y can be associate to the power consumption. PRR oes not require complex logic to be effective traffic matrix ` which escribes the actual traffic flowing between the master unit an each slave unit: ` GI ba GUI \[ e] ^f_ ORQ GI (3) I ScI T Fig.. Topologies of Bluetooth networks: piconet & scatternet ue to our assumptions of statistical routing an absence of SAR in slave to slave communications we obtain: Nml Pfn i agi k Ntl Pfn FHGUI 9+op + + ;7+qArbBXs a $I G F GUI Ar @x yzc{ +~} + ;7+uvbBXs a GI ()
Y g Q Q ^ G C _ N.. C _. _ an: h i GUI is the equivalent loa factor for the +A Pfn GI Nm K GI F GI + p + c+;7+ 9+.+ s (see M III-A for more etails). Pfn $I Nt B G K GI C Now we nee to fin the ST of the cycle time F GI Ar + H+;7+ 9+.+ s GI +A bbx+ +.+ (5) There is however a more elegant way to pass irectly from to `. et us efine the matrix as:... + (6)... where the B -th row is the only non zero one an as the all zero matrix presenting a single one in position BX+ B. Then we may write: ` + where Y Y (7) is the ientity matrix. III. PERFORMANCE ANAYSIS The elay on the link +A GUI " + PRQ ^X consists of two components the access elay (i.e. the time it takes for a packet to get to the hea of the queue an start the service) an the transmission elay! ". By using the corresponing STs (aplace-stieltes transforms) we have: $ 3 GI %& %& (8) N + where % " I SI T GI B ('). The access elay itself may be expresse as the sum of two components: -/. + (9) where. is the time elapse between the packet arrival an the first time the queue gets the token whereas the secon term takes into account the time spent for transmitting all the packets foun waiting in queue. Passing to the corresponing STs we obtain: %& %&. %& () The first term may be thought as the resiual life in a renewal process with renewal perio equal to the cycle time. Thus from the theory of ranom look [] we have for the pmf: 9;E= 3 76 8 : ; >8?A@ B7C + () BFC :G ;E= %&.?A@B C () where is the probability istribution of the r.v.. Passing to the STs we get: The other term in (9) may be foun by consiering the equivalent /(HIAH moel of the queue with an equivalent service time equal to the cycle time thus getting in terms of STs GI agi amgui %! K G ;5= + (3) where GUI a GUI a GI PRQ I SI T GUI U Pfn ]?A@ B/W a GUI GUI -th queue. The cycle perio may be expresse as the sum of the times spent for ata exchange on the +A BON -th link which ue to our assumptions are assume inepenent an thus: ;E= P I () Accoring to the notation of [6] we set I?Q@ ] ^X_ GUI. Finally consiering that in case of no ata packets available a one slot long ummy packet (PO/NU) is sent we fin: S I K GUI GI GI GI On the whole we get: loa factor of the -th queue may be escribe by: GI amgui?q@ " bagi @ v GUI GI loa factor GI GI GI GI e S GUI GUI GUI T ;ER (5) TG ;5= 3 GI?A@ BC K! G ^X_ ; = [ GI B (6) $ Equation (6) provies the complete statistics of the elay on the link +A may be foun to be:. In particular the average value 3 GI?A@ GI AU 3?A@B/W 3 GI @B C amgi?a@b C BXC may be foun by eriving expres- where the moments of sion (). Finally the average packet elay may be compute as: A. Stability Conitions Generally speaking the loa factor as []: GUI + (7) GI a GI 3 GUI 3 N (8) GI a GI of a queue is efine?q@y?a@[\ + (9) Y [ where is the service time an the arrival rate. A polling system employing PRR is stable if an only if the following conitions are both verifie []: [ GI^] 9+ () GI GI ] -+ +A + + ; () where the first conition applies to the system as a whole whereas the secon ensures that also every single queue is stable. Accoring to the notation previously introuce each. The equivalent may be calculate by consiering the equivalent / (H IAH system B C which presents a service time equal to the?q@b C service time. Thus GI amgui. Substituting after a
Q Q.5 x 5.9.8 Analytical Simulation 3.5.7 3 H H3 H5.6 G S.5 k (s).5..5.3..5..5.5.5 3 3.5.5 G M x 5 Fig. 3. Stability region in terms of offere traffic ( packets only).5 3 3.5.5 5 5.5 6 G (b/s) x 5 Fig. 5. Average packet elay versus offere traffic 7 8 x 5 7 H H+H3 H+H3+H5 N= N=3 N=5 N=7 6 G S 5 3 (ms) 8 6.5.5.5 3 3.5 G (b/s) x 5 3 5 6 7 8 G x 5 M Fig.. Stability region in terms of offere traffic few algebra we get that conition () implies () an is thus not only necessary but also sufficient to get a stable network. In Fig.3 we reporte the results for the case when only 3 76 packets are use on both forwar an reverse links for a piconet with ; slave. The achievable rate region for ; is shown in Fig.. Note that as expecte ue to the higher efficiency (in terms of payloa/packet length ratio) enabling the use of multi slot packets effectively enlarges the achievable rate region. The same reasoning applies to the general case of a piconet with ; slaves where the achievable rate region may be epicte as a polyheron in a @; imensional space. B. Channel Utilization In wireless systems banwith is a scarce resource an hence it is of vital importance to fully exploit the available banwith. The time ivision uplex (T) architecture of Bluetooth systems makes impossible to fulfil this requirement in the case of strongly asymmetric traffic. (An example may be a multicasting in a conference room where ownlink traffic only is present). Hence a metric of great interest is the so calle channel utilization parameter efine as the average percentage of slots occupie by ata packets. The channel Fig. 6. " % Average packet elay versus offere traffic for various utilization parameter which will be enote by may be obtaine by the ratio between the total number of slots use to sen ata uring a service cycle an the cycle uration expresse in slots. Hence using average values we get:?a@ B7C $[ GI C. Numerical Results GUI?Q@ " [ GI a GI?A@ " () In this section we report some results obtaine through numerical simulations. To valiate our moel we implemente a simulator using OPNET; we consiere a scenario of a full piconet (; ) with ownloa traffic only an various SAR policies. The results in terms of average packet elay are plotte in Fig.5 for S + 9+.+. In Fig.6 we reporte the impact of the number of slaves on the packet elay for various ; in a balance scenario with GUI S for any +A. In case of slave to slave communications the en to en packet elay clearly epens on the orer the master polls the slaves. To avoi such a epenency (which results in fairness loss) we shoul consier a PRR in which the polling orer is ranomize at each cycle; thus an approximate expression of the packet elay for a +A communication may be given by: 3EGUI t3egi 3 $I (3)
.35.3 Analytical Simulation.5.8 (s)..5 η.6......5.5.5.5 3 G (b/s) x 5 Fig. 7. Average packet elay for a slave to slave communication Fig. 8..6..8..3..5.6.7 π.8 k3 (3).9 π k3 (5) Channel utilization parameteras a function of " an % To show how the assumption of inepenent flows affects the results we simulate a piconet with slaves communicating each other: theoretical an simulation results are shown in Fig.7. It may be note that the assumption of inepenent flows although proviing goo results at low traffic loa leas to substantial mismatch with the simulation results as the system gets close to the stability limit. Finally we investigate the impact of the SAR policy employe: namely we analyze a piconet with ;. slaves; two of them communicate with the master at a bit rate of Kb/s using 3 packets only (which can be thought as a moel for voice over AC) whereas the thir slave is ownloaing from the master at a Kb/s. We varie the for the last connection rate of. an plotte the channel utilization parameter an the average packet elay on the voice links in Fig.8 an Fig.9 respectively. The results are worth some comments: the use of multi slot packets inee while enlarging the capacity region achieves lower efficiency in terms of banwith utilization. This result is in practice ue to the higher efficiency of multi slot packets an thus the terms in () ecreases when longer packets are use. Furthermore longer packets may result in etrimental impact on links carrying elay sensitive flows. In a QoS oriente scenario it is clear that the choice of using multi slot packets shoul be somehow negotiate at the master sie taking into account the possibly negative effects on link performances. Fig. 9. k (s).5 3.5 3.5.5.5 x 3...6 π k3 (5).8.5 π k3 (3) Average packet elay for the elay sensitive application as a function of " an % more realistic SAR policies in which the packet length is chosen in function of the buffer length an as a consequence the analysis of the packet elay at CAP level. The last interesting issue woul be the extension of such a framework to accomoate a scatternet structure where other problems (the moeling of the gateways behavior an inter piconet interference) arise. I. CONCUSION In this paper we presente a mathematical framework base on queueing theory tools for performance evaluation in a Bluetooth piconet using multi slot packets. We iscusse stability conitions an showe that the use of multi slot packets ue to their higher efficiency effectively enlarges the bounaries of the achievable capacity region. We showe how the channel utilization parameter a significant metric in wireless networks performance may be calculate an iscusse the traeoffs involve in the choice of a given SAR policy. Three subects seem to be of great interest for future work. The first is the stuy of how channel failures impact on piconet performances an their integration into a queueing theoretic framework. The secon is the effective moeling of REFERENCES [] Bluetooth special interest group. [Online]. Available: http://www.bluetooth.com []. C. Haartsen The Bluetooth raio system IEEE Pers. Comm. vol. 7 no. pp. 8 36 Feb. [3]. Haartsen M. Naghshineh. Inouye O. oeressen an W. Allen Bluetooth: ision goals an architecture ACM Mobile Computing an Communications Review vol. no. pp. 38 5 Oct 998. [] P. ohansson M. Kazantziis R. Kapoor an M. Gerla Bluetooth: an enabler for personal area networking ACM/IEEE Trans. Net. vol. 5 no. 5 pp. 8 37 Sep.-Oct.. [5] R. Bruno M. Conti an E. Gregori Bluetooth: architecture protocols an scheuling algorithms ACM Cluster Computing vol. 5 pp. 7 3 Apr. [6]. Misic an. Misic On Bluetooth piconet traffic performance in Proc. of IEEE PIMRC isbon Portugal. [7] A. Capone M. Gerla an R. Kapoor Efficient polling schemes for Bluetooth picocells in Proc. IEEE ICC Helsinki.
[8] M. Kalia. Bansal an R. Shorey ata scheuling an sar for Bluetooth mac in Proc. IEEE TC (Spring) Tokyo. [9] M. C. alenti M. Robert an. H. Ree On the throughput of Bluetooth ata transmissions in Proc. of IEEE WCNC Orlano. [] H. evy an M. Sii Polling systems: applications moeling an optimization IEEE Transactions on communications no. pp. 75 76 Oct 99. []. Kleinrock Queueing Systems. New York: ohn Wiley & Sons 975.