خروجی پیچیدگی تجارت کردن برای ترافیک ABR در یک شبکه ماهواره ای ATM تحت محدودیت های از دست دادن سلول
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|22324||2000||13 صفحه PDF||سفارش دهید||7572 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Computer Communications, Volume 23, Issue 11, 1 June 2000, Pages 1034–1046
This work refers to an ATM satellite system, supporting different QoS classes via the standard ATM traffic categories, i.e. Constant Bit Rate (CBR), Variable Bit Rate (VBR), Available Bit Rate (ABR) and Unspecified Bit Rate (UBR). The focus of this work is on ABR (i.e. closed loop control) and the central issue is to find trade-offs between a high ABR throughput and processing burden/buffer space required on board satellite, under the constraint that no buffer overflow occurs. ABR fairness issues are not central to this work and hence are not considered. The focus of the paper is to evaluate how adaptive prediction of higher priority CBR/VBR traffic can enhance ABR traffic performance, with respect to a simpler worst case based approach. Results show that a significant improvement of performance can be obtained even with rather straightforward prediction algorithms.
A large interest is currently addressed towards the convergence between Asynchronous Transfer Mode (ATM) and satellite technologies , , , ,  and . Satellite systems are inherently able to provide interconnection functions over a wide geographical area (including remote, urban and rural) and to potentially support high bandwidth communications. Further advantages can be achieved by using the broadcast capabilities and configuration flexibility of the satellite systems and the on-demand bandwidth access flexibility of ATM . A critical issue in a satellite environment is efficient exploitation of the link bandwidth, yet supporting different QoS classes. In the context of ATM, traffic management functions have been specified by ITU-T and ATM Forum  and , based on the definition of a number of service categories: Constant Bit Rate (CBR), Variable Bit Rate (VBR), Available Bit Rate (ABR) and Unspecified Bit Rate (UBR). Real time traffic can be carried by means of CBR/VBR connections; these traffic are given priority over other traffic categories, namely ABR and UBR, to meet the strict performance requirements, especially as for delay and delay jitter. However, high media utilization requires efficient exploitation of the spare bandwidth, including the one left unused by high priority CBR/VBR traffic, due to source activity (i.e. actual usage of assigned capacity may be less than 100% for a CBR/VBR connection ). ABR and UBR traffic categories lend themselves to such statistical multiplexing, but ABR should also offer some QoS guarantees, i.e. essentially lossless cell transfer (no buffer overflow) for delay insensitive data. ABR uses a feedback control loop per ABR Virtual Connection (VC) between data source and destination to control the congestion phenomena that statistical multiplexing can cause. Application of such a congestion control strategy to a satellite environment is critical due to the very large propagation delay of the satellite hop and the affordable complexity on board satellite, both as for buffer space and processing power. The specific focus of this paper is on the definition and evaluation of an ABR Explicit Rate (ER) computation algorithms, based on the exploitation of an adaptive prediction scheme to account for the real activity of CBR/VBR traffic. The ABR capacity computation algorithm is aimed at avoiding on board ABR buffer overflow. Since emphasis is on the ability of increasing ABR throughput with as minimum as possible a priori knowledge of the high priority CBR/VBR traffic and on board resource (buffer space and processing power), fairness is not a prime issue in the present context and no special consideration has been devoted to complex fair algorithms for sharing the ABR bandwidth among many sources. Motivation for this study arose in the framework of a pilot project carried out under the support of European Space Agency (ESA), including the realization of an emulation testbed of the satellite system and a field trial. The content of the paper is structured as follows. In Section 2 we briefly outline the overall system architecture and operation. In Section 3 we introduce the prediction based algorithm to evaluate the ABR allowable bandwidth. Section 4 is devoted to the performance analysis, obtained by means of simulations. Section 5 summarizes the main results of this work.
نتیجه گیری انگلیسی
The basic achievements of this work are the definition of a reactive congestion control for support of ABR via satellite that is based on an ER computation algorithm including prediction of higher priority CBR/VBR traffic. The computational burden of the algorithm is discussed and argued to be well within feasibility. To evaluate the prediction based algorithm benefits, a throughput performance evaluation is presented with a comparison of this algorithm with a simpler worst case approach, defined in Ref. . The main conclusions of this study, from a performance point of view, are: • a significant improvement of ABR throughput performance is to be expected if a minimum amount of downlink capacity is guaranteed to ABR traffic, the more, the larger this capacity; • increasing the buffer size causes a linear increase of the achieved ABR throughput, up to saturation of the whole capacity left by higher priority traffic; • the best performance are achieved when the link has a large capacity and a great number of high priority CBR/VBR sources are multiplexed; in any case, the overall downlink capacity utilization can be increased several times (at least doubled) if ABR traffic handled by the RCC is added to CBR/VBR traffic, even with very little a priori knowledge of that traffic and with the constraint of no buffer overflow for the on board ABR buffer. • since the order of the CBR/VBR traffic prediction filter, n, is equal to a few tens (n=48 at most, for the cases shown in Section 4.3) and the number of operations (flops) required by the prediction algorithm is O(n2), the processing power to be provided on board is in the order of a few hundreds of kflops/s per downlink, with a computation period as small as 10 ms; this relatively limited processing power brings about large improvements of the throughput performance with respect to the worst case assumption. Further work aims at defining a fully adaptive and robust prediction based algorithm, including use of neuro-fuzzy techniques.