Aommodations of QoS DiffServ Over IP and MPLS Networks Abdullah AlWehaibi, Anjali Agarwal, Mihael Kadoh and Ahmed ElHakeem Department of Eletrial and Computer Department de Genie Eletrique Engineering Eole de Tehnologie Superieure Conordia University Universite du Quebe 455 de Maisonneuve Blvd. W. MONTREAL, CANADA MONTREAL, Canada (HG M8) {aa_alweh@ee.onordia.a;ahmed@ee.onordia.a} Abstrat:- Multiasting has beome inreasingly important with the emergene of Internet-based appliations suh IP telephony, audio/video onferening, distributed databases and software upgrading. IP Multiasting is an effiient way to distribute information from a single soure to multiple destinations at different loations. In pratie IP is onsidered as a layer protool. Multiprotool label Swithing (MPLS) replaes the IP forwarding by a simple label lookup. MPLS ombines the flexibility of layer routing and layer swithing. In this paper, we present a new fair share poliy (FSP) that implements Differentiated Servies to solve the problems of QoS and ongestion ontrol when multiasting is used. Analysis tools are used to evaluate our new fair share poliy (FSP) for different senarios. The results should provide insights for the omparisons between IP multiast in MPLS networks using FSP and plain IP multiasting using the same poliy when DiffServ are adopted. Key-Words: Multiast, IP, MPLS, DiffServ, QoS. Introdution Multiasting has been at the enter of interest in the area of Internet ativities and has already ontributed to some major suesses. IP Multiast supports group ommuniations by enabling soures to send a single opy of a message to multiple reipients at different loations who expliitly want to reeive the information []. With the huge inrease demand for bandwidth, one of the hallenges the Internet is faing today is to keep the paket forwarding performane up. Reent developments in Multiprotool label Swithing (MPLS) open new possibilities to address some of limitations of IP systems. MPLS is an Internet Engineering Task Fore (IETF) standard []. It replaes the IP forwarding by a simple label lookup mehanism. MPLS ombines the flexibility of layer (L) routing and layer (L) swithing, whih enhanes network performane in terms of salability, omputational omplexity, lateny and ontrol message overhead. Besides this, MPLS offers a vehile for enhaned network servies suh as Quality of Servies (QoS)/ Class of Servie (CoS), Traffi Engineering and Virtual Private Networks (VPNs). IP multiast in MPLS networks is still an open issue [-4]. On the other hand, the IETF DiffServ working group is looking at a more salable model and more likely to be easier to implement than IntServ/RSVP model [5]. In the DiffServ arhiteture, traffi that requires the same Per-Hop-Behavior (PHB) is aggregated into a single queue. The DiffServ arhiteture fouses on the use of DiffServ (DS) byte, whih is the redefined 8-bit Type of Servie (TOS) field in the IPv4 header or the IPv6 Traffi Class otet as a QoS mehanism. are lassified into the orresponding queues using their DiffServ Code Points (DSCP). use DSCP bits in order to reeive a partiular PHB, or forwarding treatment. Marking, lassifiation, traffi onditioning or poliing are done at network This work was supported by a CRD grant from NSERC/BELL Canada
boundaries (first router for example) and paket treatment and handling is arried on eah network node [6]. It would be interesting to ompare QoS performane of IP and MPLS multiasting, given their partiular onstraints. In regular IP multiasting only overhead pertaining to IP multiast tree should be established, while in MPLS multiasting we have to add also the orresponding MPLS multiast tree establishment times and ontrol pakets. In this paper and taking all the above onstraints into onsideration, we evaluate the QoS performane for a typial router in the two ases of IP and MPLS multiasting. We also onsider Differentiated Servies; i.e. traffis with different priority lasses. Fair Share Poliy (FSP) FSP is not a all admission rather it is a traffi poliing mehanism. In FSP, pakets are disarded in ase of ongestion differently at eah queue aording to soure priority and the maximum number in the queue; i.e. the soure with higher priority will experiene less paket disarding than soures with lower priorities. FSP guarantees fairness among flows having the same priority (i.e., required QoS) in two respets: Firstly the buffer spae alloated to lower priority traffi is larger; thus leading to less paket disard. Seondly by seletive paket disarding on pakets from the same flow and making sure that the total number of pakets disarded per flow is the same for all flows with the same QOS requirement. In this paper, we only explore the first fairness mehanism. The Analytial Model Our analytial model is shown in Fig.. In this model, a typial IP or MPLS router and our FSP traffi poliing mehanism proess three independent soures orresponding to different input traffi lasses. Soure is assigned the highest priority, then soure and finally soure. For this model, the enforement is assumed to our at the router (node) aording to Fair Share Poliy. The following assumptions are used: - Assume a Bernoulli arrival for all soures; in order to be short and disrete inter-arrivals. - FSP uses non pre-emptive priority queuing. - The arrival probabilities are, and for eah soure respetively. Note that represents the probability of reeiving a paket while one paket is served on the hannel. 4- Servie disiplines for different queues are β, β and β for eah soure respetively. 5- Average queue sizes are n, n and n for eah soure respetively. 6- Maximum buffer sizes are nmax, n max and n max for eah soure respetively. 7- Total system buffer size: B = n max + n max + n max, where n imax and i =,, is alulated as: n n n max = (( )*B), n (( max= *B)) n + n + n n + n + n n and n max = (( )*B) n+ n+ n 8- All of MPLS or IP routers on the subjet Internet are homogeneous in providing resoure and traffi onditions, so we take one of them as a representive for IP routers and the other one as a representaive for MPLS routers. 9- All pakets are of the same length. Soure Soure Soure Arriving Dropped Arriving Dropped Arriving Dropped Input Buffer FSP Input Buffer FSP Input Buffer FSP Priority assignment. for Soure. for Soure for Soure Fig. The analytial model Node Departing P
. The Coupled State Diagrams The oupled state diagrams for the analytial model in Fig. are shown in Fig.. This diagram represents a typial router with priority lasses. The solution of the number in every lass depends on the solutions of the other lasses.; where β = always in order to give soure with highest priority the best servie probability, β ; i.e. pakets from = P soure will be served only when the buffer orresponding to soure (whih has higher priority) is empty and finally β P = P ; i.e. pakets from soure top of IP layer), is the extra arrival probability due to IP ontrol overhead used to establish the IP multiast tree. The above equation an be rewritten in terms of as: = τ + = where is where is a fator. Similarly for MPLS based networks, an be written as: = + + Where and are the same as in the ase of IP networks; is the extra arrival probability due MPLS ontrol overhead used to establish MPLS multiast paths or tree. an be rewritten in terms of as: ( = + + ) = where is a fator.. Solution of the Analytial Model By writing the balane equations for the state diagram in Fig. [7], and solving these equations by iteration to find the probabilities. In order to write the equations in simpler forms we define: λ = ( P, β) = ( )P β and σ = P β + ( )( P β) () Fig. The Coupled State Diagram will be served only when the buffers orresponding to soure and soure (whih have higher priority) are all empty. P is the probability that reflets the omulative effets of hannel error and ongestion of next router (i.e. the probability of no loss or errors on the hannel). Paket loss probability for eah soure an be obtained by alulating the probability to be in last stage in the state diagram Pnmax, Pnmax and P nmax respetively. For IP based networks, the soure arrival probability is atually a ompsite one; for instane (for soure ) an be written as: + = τ +, τ = Where is the proessing time at lower layers (for example MAC layer) and is the proessing time at IP layer, is the intrinsi arrival probability at the appliation layer (on To find a speifi probability P i : σ λ P i = P i P i Where i=,4 nmax. () Notie that P = ( ) P () σ and P = P P (4) P i an be rewritten in the following form: P i = (ai + bi ) P (5) where i=,4 nmax; and nmax is a speifi soure maximum buffer size. σ λ ai = ( ai ai ) (6) ( σ) Where a = a = and a = σ λ b ( b b ) (7) i = i i λ Where b = b = b = and b = nmax Taking into aount that: P i = i =
P an be found using the following: P = nmax (8) + (a i + b i ) i = then any probability an be found in terms of P as in equation (5). The average number of pakets in the buffer for a speifi soure an be found as: n = nmax i * P i (9) i = Notie that the loss probability is equal to the probability to be in last stage of state diagram; for example the loss probability for soure is: P L = P nmax. 4. Analysis Results Fig. shows the average number of pakets in the system buffer versus for all soures for both IP and MPLS. The figure shows that IP and MPLS will have very similar average number of pakets espeially for low priority traffi and when the intrinsi arrival rates are relatively high. Note that the value of τ is relatively small in whih we assumed that the differene in paket proessing between IP and MPLS is small. Average number of pakets 8 6 4 Average number of pakets in the buffer for all soures for both IP and MPLS,5,,5,,5, =., =., =.5,β =, P =.8, B =, =., τ =. Fig. Average number of pakets in the buffer for all soures for both IP and MPLS (small τ) Fig.4 shows the paket loss probability for all soures for both IP and MPLS versus for relatively high intrinsi arrival rates and small τ. It shows that IP and MPLS have almost the same loss probability, IP IP MPLS MPLS exept a small differene for soure ; and as inreases the differene beomes even smaller. Paket loss probability,9,8,7,6,5,4,,, Paket loss probability for all soures for both IP and MPLS,5,,5,,5, Fig.4 Paket loss probability for all soures for both IP and MPLS (small τ) However, Figs.5 and 6 show that when τ inreases MPLS will have superiority over IP in terms of average number of pakets in the system buffer and paket loss probability. As shown in Fig. 5, the average number of pakets in the system buffer in the ase of MPLS is less than IP for all soures and this differene is lear for low priority soures and. Fig. 6 shows that the paket loss probability in the ase of MPLS is less than IP for all soures. This means when the differene in paket proessing τ between MPLS and IP inreases, MPLS will be better. Average number of pakets Fig.5 Average number of pakets in the buffer for all soures for both IP and MPLS (large τ) IP IP MPLS MPLS =., =., =.5,β =, P =.8, B =, =., τ =. 8 6 4 Average number of pakets in the buffer for all soures for both IP and MPLS,5,,5,,5, IP IP MPLS MPLS =., =., =.5, β =, P =.8, B =, =., τ =.8
Paket loss probability,,8,6,4, Paket loss probability for all soures for both IP and MPLS,5,,5,,5, Fig.6 Paket loss probability for all soures for both IP and MPLS (large τ) In the previous figures, 4, 5 and 6 was onstant and relatively small; that s why MPLS performane was better or very similar to IP performane. However, in the next figures we will study the effets of on MPLS performane. Figs. 7 and 8 show that IP will have superiority over MPLS when inreases espeially for soures and. As shown in Fig.7, the average number of pakets in the system buffer in the ase of IP (whih is onstant) is less than MPLS. Average number of pakets =., =., =.5, β =, P =.8, B =, =., τ=.8 8 6 4 Average number of pakets in the buffer for all soures for both IP and MPLS,,4,6,8,, Fig.7 Average number of pakets for all soures for both IP and MPLS (Effet of ) Similarly, Fig. 8 shows that paket loss probability in the ase of IP (whih is IP IP MPLS MPLS IP IP MPLS MPLS =., =.5, =., β =, P =.8, B =, =., τ =. onstant) is less than MPLS. This means when the extra arrival rate due MPLS ontrol overhead used to establish MPLS multiast paths or tree inreases, IP will be perform better. Paket loss probability Paket loss probability for all soures for both IP and MPLS,,8,6,4,,,4,6,8,, IP IP MPLS MPLS Fig.8 Paket loss probability for all soures for both IP and MPLS (effet of ) 5. Conlusions and Future Work In this paper, a omparison between IP multiast sessions and MPLS multiast sessions is arried using analysis tools. In addition to that a new Fair Share Poliy (FSP), whih is a traffi poliing mehanism is proposed to ensure proper QoS. Also, Differentiated Servies are used in this omparison. In this paper, we found that when the differene in paket proessing time between IP and MPLS is high, IP multiast will perform less effiiently than MPLS in terms of average number of pakets in the system buffer and loss probability. However, when this differene is small, IP performs very similar to MPLS. In addition to that when MPLS have higher arrival rate due MPLS trees establishment ontrol overhead, it would perform worst than IP. Taking the same values of and for all priority lasses; impliitly assumes a shared tree. In the near future different values for and for different priority lasses (soure trees) an be assumed. Also, in the oming future, ARQ/FEC an be implemented in our analytial model or similar model to ensure a guaranteed delivery of multiast pakets. =., =.5, =., β =, P =.8, B =, =., τ =.
Referenes: [] B. Quinn et all, " IP Multiast Appliations: Challenges and Solutions ", RFC 7, September ; http://www.ietf.org/rf/rf7.txt. [] E. Rosen, A. Viswanathan and R. Callon," Multiprotool Label Swithing Arhiteture ", RFC, January ; http://www.ietf.org/rf/rf.txt [] D. Ooms et all, " Framework for IP multiast in MPLS ", IETF Draft, draft-ietfmpls-multiast-7.txt, January. [4] D. Ooms and W. Livens, " IP Multiast in MPLS Networks ", Proeedings of the IEEE Conferene on High Performane Swithing and Routing, June, pp. 5. [5] J. Wroklawski, " The Use of RSVP with IETF Integrated Servies ", RFC, September 997; http://www.ietf.org/rf/rf.txt [6] M. Carlson et al., " An Arhiteure for Differentiated Servies ", RFC 475, Deember 998; http://www.ietf.org/rf/rf475.txt [7] T. Saadawi, M. Ammar and A. Elhakeem, " Fundamentals of Teleommuniation Networks ", New York, NY: John Wiley & Sons In