Performance analysis of the simple prioritized buffering algorithm in optical packet switch for DiffServ assured forwarding

Performance analysis of the simple prioritized buffering algorithm in optical packet switch for DiffServ assured forwarding
of 4
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
      This work has been partially funded by the Catalan Government through the International Graduate School of Catalonia grant and MCYT (Spanish Ministry of Science and Technology) under contract FEDER-TIC2002-04334-C02-02. Performance Analysis of the Simple Prioritized Buffering Algorithm in Optical Packet Switch for DiffServ Assured Forwarding Miros ł aw Klinkowski (1,2) , Davide Careglio (2) , Marian Marciniak  (1) , Josep Sole-Pareta (2)   (1)  National Institute of Telecommunications, Department of Transmission and Fiber Technology,1 Szachowa Street, 04-894 Warsaw, Poland e’mails: M.Klinkowski@itl.waw.pl, M.Marciniak@itl.waw.pl  (2) Universitat Politècnica de Catalunya (UPC), Advanced Broadband Communications Centre (CCABA) Jordi Girona, 1-3, Mòdul D6 (Campus Nord), 08034 Barcelona, Catalunya, Spain e-mails: {mklinkow, pareta, careglio}@ac.upc.es   ABSTRACT This paper addresses the problem of providing QoS management for optical packet switch. In particular we  propose the Simple Prioritized Buffering Algorithm (SPBA) for prioritized buffer’s management, which aims to guarantee different packet loss probabilities to different packet streams. In contrast to other solutions, SPBA does not need any resource reservation or threshold dropping, but only makes use of priority scheduling. A theoretical model of SPBA with  K   classes of services applied to an optical packet switch with fiber-delay lines  buffers and tunable wavelength converters is presented. The analytical model is verified by simulations. Keywords : Optical Packet Networks, Services Differentiation, Assured Forwarding, Performance Analysis, QoS, Buffering Management Algorithms 1.   INTRODUCTION Packet loss is not uncommon in packet networks. As the network, or even some of its links and nodes, becomes congested, router buffers fill and start to drop packets. For non-real-time applications, such as file transfer and e-mail, packet loss is not critical. In real-time applications (i.e. voice, video, telemedicine services) packets lost would cause clipping or unintelligible speech as well as discontinuous picture. Although transmission in optical  packet networks is very fast and in general packets delay could be lower than in electronic packet networks (because of transparent optical packet transmission without O/E conversion in intermediate nodes), these applications would still demand better packet delivery guarantees. In this instance some techniques are necessary to improve performance of the optical packet node for some application regions. Services differentiation that is  beginning to play important role in traditional core packet networks with big traffic load has been proposed as the solution of QoS for optical packet networks [BSH02, HM0202]. In this paper we present a multi-class contention resolution algorithm, called Simple Prioritized Buffering Algorithm (SPBA), to provide service differentiation that can be useful for Assured Forwarding (AF) implementation in an optical packet switch. A theoretical model of SPBA with  K   classes of services is discussed and the goodness of the model is verified by simulations.  2.   SERVICES DIFFERENTIATION IN OPTICAL PACKET SWITCH In this paper we assume the availability of an optical packet switch with O/E conversion for the packet headers to take routing and forwarding decisions and transparently switching for packet payloads. We consider fixed-length packets with synchronous operation. Packets compete the same output can use the time domain through a  pool of fiber delay lines or wavelength domain through wavelength multiplexing and wavelength conversion. Fig. 1 illustrates the optical packet node and buffer’s model. 2.1   Overview Services differentiation addresses the clear need for relatively simple and coarse methods of categorizing traffic into different classes of priority, also called class of service (CoS), and applying QoS parameters to those classes. To accomplish this, packets are first divided into classes by marking the type of service field in the  packet header. Once packets are classified at the edge of the network, specific forwarding treatments are applied on each network node. This combination of packet marking and well-defined servicing procedure results in a scalable QoS solution for any given packet, and thus any application. Services differentiation adopts the core network to provide a form of QoS treated from the aggregate perspective. Compared to other solutions based on   bandwidth reservation, signaling is eliminated, and the number of states required to be kept at each network element is drastically reduced, resulting in a coarse-grained, scalable and end-to-end QoS solution. Therefore services differentiation QoS approach is appropriate for the core optical packet networks where accurate QoS metrics for individual flows could not be assured because of difficulty in selective packets control (storing, switching) in optical nodes but aggregated servicing could be well performed. λ 1... λ w   λ 1... λ w   1    N   DMUX   λ 1   λ w   λ 1   λ w   TWC   Switch Matrix   Buffers (of size B)   λ 1... λ w   λ 1... λ w   1    N   B/w + 1 inputs   Buffers (of size B) B = w * D 1   w   w   a k   a) Model of the Photonic Node   b) Model of the Buffer    1xT2xTDxT    Figure 1. The optical packet node and the buffer 2.2   Buffer management algorithm The incoming packets are dropped in optical packet node when all wavelength channels of the destined node outlet are busy and there are not resources to store (delay) or deflect them at the moment. Some traffic with higher priority class could need better guarantees in respect of maximum packet loss rate. Under high traffic load in the node too much priority packets would be dropped, so some control decision algorithms and resource guarantees are needed [KM03]. The simplest one seems to be SPBA algorithm. The task of the SPBA is to store into the buffer all incoming  packets starting with the highest priority packets. If there are not free slots available remaining packets are dropped. The SPBA in contrast to PBSO [HM0202] and PBS [HM0207] algorithms does not apply any resource reservation mechanisms and therefore cannot be dynamically tuned to the temporary traffic profiles. However  both its simplicity and lack of the necessity for enhanced node architecture and lack of a need for calculating the optimal thresholds in PBS and PBSO address the SPBA to be an interesting option for prioritized buffering management in the optical node. 3.   ANALYTICAL MODEL OF THE SPBA ALGORITHM We use following notations to represent the system behavior:  N   – the number of input/output ports w  – the number of wavelength channels in the input/output ports  D  – the number of fiber delay lines in the buffer; each delay line generates an adequate delay T, 2T, … BT    T   – the duration of the time slot equivalent to the length of a packet  B  – the buffer capacity, equals the number of delay lines D increased in number of wavelengths w    K   – the number of priority classes (1 is the higest, K is the lowest)  ρ   – the traffic load rate per channel  ρ  1  ,  ρ  2  , …,  ρ   K    denote the ratio class-1, class-2, …, class-K packets to the total number of packets; 1 1 = ∑ =  K ii  ρ   For the analyzed system it is assumed that the packets arrive synchronously and all packets have the same fixed length that corresponds to the time slot (packet duration = time slot). In each time slot between 0 and wN  packets have to be distributed to the  N output ports. The time independent load  ρ    gives the probability of a packet arriving at one of the w  wavelength channels per input port. Assuming this, probability a k   that i  packets arrive at buffer in a slot: iwN ii  N  N iwN a −     −        =  ρ  ρ  1 , where 0 ≤   i   ≤   wN   (1) Probability that n 1  class-1 packets, n 2  class-2 packets, … n  K   class-  K   packets arrive at the system in a time slot, since  K  priority classes are considered is given by:    ∏∑ ×∑= iiiin K nnnnnn nnab  K ii K  )!()!( 2121 21,,  ρ  ρ  ρ   K K , (2) where  ρ  1  ,  ρ  2  , …,  ρ   K    mean the ratio class-1, class-2, …, class-  K   packets to the total number of packets.   Let Q  be a random variable that indicates how many packets are stored in the buffer. The corresponding steady-state probability of the queuing system is denoted as ),( iQ P q i ==  where i = 0, …, B . The queue is characterized by the number of packets that can be stored  B  and by the number of packets that can leave the queue in each time slot w . The matrix of state transition probabilities  P  ij , that represents the transition  probability of having  j  packets in the buffer at the end of time slot m  given there were i  packets in the buffer at the end of time slot m-1 , for analyzed system have been studied in [DMS97]. The steady-state probability of Q can be found by solving stationary equation ' Q P Q =⋅ .  Using calculated transition probabilities, the Packet Loss Rrate (  PLR ) can be found as 1 – PSR , where  PSR is the Packet Success Rate. Taking into account considered prioritized buffering algorithm, Packet loss rates for the individual priority class  p  can be obtained as ∑∑∑∑∑ −==−=−== −+−∑−= −= 11000,,0 )),,0min(max( 11 1121121  pk  pk wN nnwN nnwN nnnn Bii p p nnwi Bbq PLR  K  j j K  K  K K  ρρ    (3) where: 1 ≤    p   ≤    K  ;  p  = 1 is the highest priority and  p  =  K   is the lowest priority class. 4.   PERFORMANCE OF THE SPBA The analytical model is verified by using simulations (Fig. 2a). The performance characteristics presented below are regarding a scenario with two priority-class of the traffic (  K = 2 ): a High-Priority (HP) traffic class and a Low-Priority (LP) traffic class. We consider a 16×16 switching matrix, with uniform traffic pattern, and mean load per input port of 0.8 Erlang.  Figure 2 . Packet Loss Rate as a function of HP traffic load, considering 5-packets buffer capacity a) comparing analytical results and simulation results with 1 wavelength, and 5 delay lines b) using different number of wavelengths (w) and delay lines (D) Figure 2a shows the PLR as a function of the HP traffic load  ρ  HP  , comparing the analytical (solid lines) and the simulation results (square markers), and assuming 5 FDLs, each one with 1 wavelength. Obtained results verify the goodness of the analytical model. We obtained improved PLR characteristics of HP traffic (Fig. 2a) that are  better since HP traffic load parameter  ρ  HP   is lower. It can be well explained when we look at the function of  buffering algorithm that store always the same quantity of HP packets independently of the  ρ  HP    rate. Therefore for smaller  ρ  HP   values we are getting better PLR characteristics. Figure 2b depicts the PLR as a function of the HP traffic load  ρ  HP  , with 5-packet buffer capacity and considering different number of wavelengths w . This figure shows that increasing number of wavelengths (which allows to  better exploit the wavelength domain), the PLR considerably decreases. Figure 2b also shows that more number of wavelengths applied in the buffer, when the same buffer capacity is kept, improve PLR characteristics for HP traffic. This is because more packets leave the buffer at the same time and there are more free slots in the buffer  for HP packets in each following time slot. The conclusion is that for a scenario with more number of low rate HP traffic flows, like i.e. multimedia streams, more wavelengths in the buffer than the number of delay lines is  preferred in order to improve PLR characteristics for this traffic. The number of delay lines has not impact on improving of PLR characteristics for HP traffic in reference to the LP traffic (Figure 3). Another remark is that for small HP traffic load  ρ  HP  , the difference between PLR of LP traffic and the scenario without priority classes is very close.  Figure 3. PLR as a function of number of delay lines, with    ρ  HP   = 0.1 HP traffic load and w = 1 wavelength   5.   CONCLUSIONS Services Differentiation is suggested QoS approach for optical packet networks since a number of services like real-time applications need some better guarantees (PLR < 10 -6 ). It offers aggregated servicing of the traffic that can be well performed especially in core optical packet networks with small buffering resources under big traffic load. A specific of optical packet network where no optical RAM’s are available yet requires of using proper control algorithm for fiber buffers’ management. Analytical study of the Simple Prioritized Buffering Algorithm (SPBA) presented in the paper showed improved  packet loss rate (PLR) characteristics, for high priority traffic with low load, without any serious degradation of PLR for low priority traffic. More wavelengths employed with less number of fiber delay lines while the same  buffer capacity is kept improves PLR rate of high priority traffic. The SPBA is not very flexible and optimal for buffers dimensioning application. Using resource reservation mechanisms based on dynamic thresholds may decrease the PLR since the buffering resources are being more  precisely tuned to the actual traffic profile. In the other hand SPBA seems to be the simplest buffering management algorithm what is especially desired for controlling the process of packets routing in optical domain. It can be applied for any number of DiffServ AF priority-classes without additional demands on the node architecture. The SPBA is especially suggested for the optical core networks if an admission control function is implemented on the edge of the network in order to balance the load characteristics of incoming traffic for each of priority class. This is a preliminary step in analytical study of prioritized buffering algorithms. The next tasks will be an investigation of optimal parameters configuration for thresholds based prioritized buffering algorithms and the introduction of a dynamic thresholds adaptation mechanism for better accommodation of traffic variations. REFERENCES [BSH02] S. Bjornstad, H. Stol, D.R. Hjelme, Quality of service in optical packet switched DWDM transport networks , Proceedings of SPIE vol.4910, APOC 2002, Shanghai (China), October 2002. [DMS97] S.L. Danielsen, B. Mikkelsen, C. Joergensen, T. Durhuus and K.E. Stubkjaer, WDM Packet Switch  Architectures and Analysis of the Influence of Tuneable Wavelength Converters on the  Performance , IEEE/OSA Journal of Lightwave Technology, vol.15, no.2, February 1997. [KM03] M. Klinkowski, M. Marciniak, Services Differentiation in MPLS Photonic Packet Networks , Proceedings of the 7th IFIP ONDM, Budapest (Hungary), February 2003.   [HM0202] H. Harai, M. Murata,  Prioritized Buffer Management in Photonic Packet Switches for DiffServ  Assured Forwarding  , Proceedings of the 6th IFIP ONDM, Torino (Italy), February 2002. [HM0207] H. Harai, M. Murata,  Performance Analysis of Prioritized Buffer Management in Photonic Packet Switches for DiffServ Assured Forwarding  , Proceedings of SPIE vol.4874, Boston (USA), pp.298-309, July 2002.
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks