Chapter 6: CALL ADMISSION CONTROL 203 Multimedia GEO Satellite Network”, in Proc. of the IEEE International Conference on Communications 2003 (ICC2003), Vol. 1, pp. 422-427, May 11-15, 2003. [28] F. De Rango, M. Tropea, S. Marano, “Controlled Load Service Management in Int-Serv Satellite Access Networks”, in Proc. of the Canadian Conference on Electrical and Computer Engineering 2004, Vol. 4, pp. 2193-2196, May 2-5, 2004. [29] C. H. Chang, H. K. Wu, Y. O. Tseng, “Quality of Service Support for Broadband Satellite Multimedia Service”, in Proc. of the IEEE Wireless Communications and Networking Conference 1999 (WCNC 1999), Vol. 1, pp. 187-192, September 21-24, 1999. [30] F. De Rango, M. Tropea, P. Fazio, S. Marano, “Call Admission Control with Statistical Multiplexing for Aggregate MPEG Traffic in a DVB-RCS Satellite Network”, in Proc. of IEEE Global Telecommunications Conference 2005 (GLOBECOM 2005), St. Louis, MO, USA, Vol. 6, pp. 3231-3236, November 28 - December 2, 2005. [31] P. Pace, G. Aloi, S. Marano, “Efficient Real-Time Multimedia Connections Handling over DVB-RCS Satellite System”, in Proc. of the IEEE Global Telecommunications Conference 2004 (GLOBECOM 2004), Dallas, Texas, USA, pp. 2722-2727, November 29-December 3, 2004. [32] Y. Qian, R. Q. Hu, C. Rosenberg, “Integrated Connection Admission Control and Bandwidth on Demand Algorithm for a Broadband Satellite Network with Heterogeneous Traffic”, IEICE Transactions on Communications, Vol. E89-B, No. 3, pp. 895-905, March 2006. [33] A. C. Fu, E. Modiano, J. N. Tsitsiklis, “Optimal Energy Allocation and Admission Control for Communications Satellites”, IEEE/ACM Transactions on Networking, Vol. 11, No. 3, pp. 488-500, June 2003. [34] J. M. S´anchez, M. Albani, C. Arbid, M. Servilio, S. H. Oh, G. Leoleis, F. Del Sorbo, G. Lombardi, A. Giralda, “Resource Management in Integrated S-T- UMTS Networks”, in Proc. of the First International Conference on Advanced Satellite Mobile Systems (ASMS 2003), Frascati, Italy, July 10-11, 2003. [35] IST FUTURE Project, Deliverable D03.03: Description and Functional design of the QoS procedure for the final UMTS, November 2002. [36] G. Matyagina, N. Shenoy, J. Asenstorfer, “Call Admission Control for a CDMA/TDMA based ATM Satellite Access Network”, in Proc. of the International Conference on Telecommunications (ICT’99), Cheju, Korea, June 15-18, 1999. [37] E. Del Re, R. Fantacci, G. Giambene, “Efficient Dynamic Channel Allocation Techniques with Handover Queuing for Mobile Satellite Networks”, IEEE Journal on Selected Areas in Communications, Vol. 13, No. 2, pp. 397-405, February 1995. [38] E. Del Re, R. Fantacci, G. Giambene, “Handover Queuing Strategies with Dynamic and Fixed Channel Allocation Techniques in Low Earth Orbit Mobile Satellite Systems”, IEEE Transactions on Communications,Vol.47,No.1,pp. 89-92, January 1999. [39] I. Mertzanis, R. Tafazolli, B. G. Evans, “Connection Admission Control Strategy and Routing Considerations in Multimedia (Non-GEO) Satellite Networks”, in Proc. of IEEE Vehicular Technology Conference (VTC 1997), Phoenix, AZ, USA, pp. 431-435, May 4-7, 1997. 204 Stylianos Karapantazis, Petia Todorova [40] S. Cho, “Adaptive Dynamic Channel Allocation Scheme for Spotbeam Handover in LEO Satellite Networks”, in Proc. of IEEE VTC 2000,Boston, USA, pp. 1925-1929, September 25-28, 2000. [41] S. Cho, Ian F. Akyildiz, M. D. Bender, H. Uzunalioglu, “A New Connection Admission Control for Spotbeam Handover in LEO Satellite Networks”, Wireless Networks, Vol. 8, No. 4, pp. 403-415, July 2002. [42] E. Del Re, R. Fantacci, G. Giambene, “Different Queuing Policies for Handover Requests in Low Earth Orbit Mobile Satellite Systems”, IEEE Transactions on Vehicular Technology, Vol. 48, No. 2, pp. 448-458, March 1999. [43] G. Maral, J. Restrepo, E. Del Re, R. Fantacci, G. Giambene, “Performance Analysis for a Guaranteed Handover Service in a LEO Constellation with a Satellite-Fixed Cell System”, IEEE Transactions on Vehicular Technology, Vol. 47, No. 4, pp. 1200-1214, November 1998. [44] L. Boukhatem, A. L. Beylot, D. Ga¨ıti, G. Pujolle, “TCRA: A Time-based Channel Reservation Scheme for Handover Requests in LEO Satellite Systems”, International Journal of Satellite Communications and Networking, Vol. 21, No. 2, pp. 227-240, March/April 2003. [45] E. Papapetrou, F. N. Pavlidou, “QoS Handover Management in LEO/MEO Satellite Systems”, Wireless Personal Communications, Vol. 24, No. 2, pp. 189- 204, January 2003. [46] E. Papapetrou, F. N. Pavlidou, “Analytic Study of Doppler-Based Handover Management in LEO Satellite Systems”, IEEE Trans. Aerosp. Electron. Syst., Vol. 41, No. 3, pp. 830-839, July 2005. [47] P. Todorova, S. Olariu, H. N. Nguyen, “A Lightweight Call Admission and Handover Management Scheme for LEO Satellite Networks”, in Proc.ofthe Fifth European Workshop on Mobile/Personal Satcoms (EMPS 2002), Baveno, Italy, Sept. 25-26, 2002. [48] P. Todorova, S. Olariu, H. N. Nguyen, “A Two-Cell-Lookahead Call Admission and Handoff Management Scheme for Multimedia LEO Satellite Networks”, in Proc. of the 36 th Annual Hawaii International Conference on System Sciences (HICSS-36), Big Island of Hawaii, USA, Jan. 6-9, 2003. [49] S. Olariu, S. R. Ali Rizvi, R. Shirhatti, P. Todorova , “Q-Win - A New Admission and Handoff Management Scheme for Multimedia LEO Satellite Networks”, Telecommunications Systems, Vol. 22, No. 1-4, pp. 151-168, January-April 2003. [50] J. Siwko, I. Rubin, “Call Admission Control for Capacity-Varying Networks”, Telecommunication Systems, Vol. 16, No. 1-2, pp. 15-40, January 2001. [51] J. Siwko, I. Rubin, “Connection Admission Control for Capacity-Varying Networks with Stochastic Capacity Change Times”, IEEE/ACM Trans. on Networking, Vol. 9, No. 3, pp. 351-360, June 2001. [52] P. Boedhihartono, G. Maral, “Evaluation of the Guaranteed Handover Algo- rithm in Satellite Constellations Requiring Mutual Visibility”, International Journal of Satellite Communications and Networking, Vol. 21, No. 2, pp. 163-182, March/April 2003. [53] E. Papapetrou, S. Karapantazis, G. Dimitriadis, F. N. Pavlidou, “Satellite Handover Techniques for LEO Networks”, International Journal of Satellite Communications and Networking, Vol. 22, No. 2, pp. 231-245, March/April 2004. Chapter 6: CALL ADMISSION CONTROL 205 [54] S. Karapantazis, F. N. Pavlidou, “Design Issues and QoS Handover Manage- ment for Broadband LEO Satellite Systems”, IEE Proc. Communications, Vol. 152, No. 6, pp. 1006-1014, December 2005. [55] S. Karapantazis, P. Todorova, F. N. Pavlidou, “On Call Admission Control and Handover Management in Multimedia LEO Satellite Systems”, in Proc.ofthe 23 rd AIAA ICSSC 2005, Italy, Rome, September 25-28, 2005. [56] S. Karapantazis, P. Todorova, F. N. Pavlidou, “On Bandwidth and Inter-Satellite Handover Management in Multimedia LEO Satellite Systems”, in Proc. of ASMS 2006, Munich, Germany, May 29-31, 2006. 7 DYNAMIC BANDWIDTH ALLOCATION Editors: Tommaso Pecorella 1 , Giada Mennuti 1 Contributors: Nedo Celandroni 2 , Franco Davoli 3 , Erina Ferro 2 , Alberto Gotta 2 , Stylianos Karapantazis 4 , Giada Mennuti 1 , Antoni Morell 5 , Tommaso Pecorella 1 , Gonzalo Seco Granados 5 , Petia Todorova 6 ,Mar´ıa ´ Angeles V´azquez Castro 5 1 CNIT - University of Florence, Italy 2 CNR-ISTI - Research Area of Pisa, Italy 3 CNIT - University of Genoa, Italy 4 AUTh - Aristotle University of Thessaloniki, Greece 5 UAB - Universitat Aut´onoma de Barcelona, Spain 6 FhI - Fraunhofer Institute - FOKUS, Berlin, Germany 7.1 Dynamic bandwidth allocation: problem definition Some of the appealing advantages of satellite networks, such as the wide coverage and the configuration flexibility, make them an ideal candidate for providing multimedia services worldwide. Satellite bandwidth is, however, a commodity at a premium, and an inefficient utilization of it may negate some of the aforementioned advantages. To this end, an apportioning scheme that dynamically allocates the bandwidth among the satellite terminals, while 208 Tommaso Pecorella, Giada Mennuti fulfilling the QoS requirements, is of paramount importance. Moreover, the satellite scenario adds a new dimension to the treatment of bandwidth, owing to the presence of both variable physical channel operating conditions and large bandwidth-delay products. Typically, control actions in telecommuni- cation networks need to be exerted over a wide range of time scales to cope with events that may occur with frequencies ranging from milliseconds to minutes or hours [1]-[4]. Satellite systems not only experience variable-load multimedia traffic, but also variable channel conditions and large propagation delays. The variability in operating conditions is due both to changes in the traffic loads and to the signal attenuation on the satellite links due to the degradations that result from atmospheric events, which particularly affect, for example, the transmissions in the Ka band (20-30 GHz). The variability due to changing radio channel conditions can be counteracted by means of Adaptive Coding and Modulation (ACM) techniques that, however, modify the available bandwidth for higher layers, thus affecting Dynamic Bandwidth Allocation (DBA) schemes. Efficient bandwidth utilization and QoS provisioning are, unfortunately, two competing goals; therefore, DBA schemes seek for a trade-off between them. To address the vast majority of IP traffic, which is inherently bursty, a technique that implicitly evaluates the bandwidth requirement at each satellite terminal and manages the traffic flows is essential. The purpose of this Chapter is to present a number of solutions for assigning the satellite bandwidth to different users (Earth stations) and traffic types. The combined action among various protocol layers (from the physical layer up to the application layer) is likely to be a good way to combat channel variability. However, this procedure could be too complex to obtain in the widest possible extent, which would imply numerous cross-layer interactions for control purposes and the related exchange of signaling information. In or- der to obtain optimized policies for satellite bandwidth allocation, the actions taken in a satellite network at the physical layer (where fade countermeasure techniques are applied) can be combined with actions at the data link layer (where the satellite bandwidth is allocated), thus realizing a more limited cross-layer optimization. The complexity of this procedure lies in the fast changing measurements required at the physical layer, regarding the channel state (signal power-to-noise ratio), which might produce an unstable allocation at the data link layer. Hence, the feedback information has to be properly filtered, possibly with some hysteresis to obtain a stable allocation at the data link layer. Regarding resource allocation, another problem is the control network architecture, which can be centralized or distributed. A centralized allocation is performed by a station, which plays the role of master (or Network Control Center, NCC). The master station collects all information relevant to the other stations (slave stations) and performs the best bandwidth allocation. This may produce a heavy computational effort in the master station. A distributed allocation technique solves the computational problem, but requires a robust Chapter 7: DYNAMIC BANDWIDTH ALLOCATION 209 control channel and an efficient control protocol, which takes into account the large communication delay. As a consequence, the available bandwidth may be significantly reduced by the signaling protocol. It should be observed, however, that the bandwidth allocation problem is somewhat different for different satellite network topologies. While the typical focus for GEO satellites is the efficient bandwidth assignment among terres- trial gateways, for LEO satellites handover and call prioritization procedures become crucial aspects. In a GEO satellite system the main limit is the time delay, in a LEO satellite system this issue is mitigated, but the system complexity causes several problems. In order to achieve a continuous satellite access, a large network of LEO satellites is required with regular handovers among them. Achieving ubiquitous coverage poses a significant challenge, and the speed at which the satellites ground track moves on the Earth generates rapidly chang- ing communication channels, subject to severe Doppler spreading. Moreover, if a constellation of LEO satellites is designed to provide global coverage, then these satellites must be able to communicate one to another, either by incorporating Inter-Satellite Links (ISLs) or a ground-based hub station in each footprint. All these issues contribute in making DBA an essential approach for providing the proper QoS but, at the same time, make its design very difficult. A less treated problem, moreover, could arise from satellite-based mesh architectures. So far, the system model only considers the uplink part, relying on the assumption that downlink is not a bottleneck. In a meshed architecture with multiple, limited-bandwidth downlink spot-beams, the channel allocation will have to take into account also this aspect in order to maintain the overall QoS; this is particularly important in satellite-based switching systems [5]. 7.1.1 Survey of allocation approaches DBA schemes can be distinguished as static and adaptive. Static algorithms In static schemes, once a terminal is assigned a certain amount of capacity, this capacity remains constant for the connection’s lifetime. The terminal can locally handle dynamically the bandwidth, without involving the NCC. That is, the assigned capacity can be apportioned between High-Priority (HP) and Low-Priority (LP) traffic. Adaptive algorithms In the case of adaptive schemes, each satellite terminal can send requests to the NCC in order to reserve or release channel capacity, based on its dynamic 210 Tommaso Pecorella, Giada Mennuti estimation of bandwidth needs. To meet the QoS requirements of bursty and delay-sensitive traffic, the terminal can follow three approaches: • Fixed allocation proportional to the maximum source rate, to be requested on a per-connection basis, • Fixed allocation at a given rate using DBA for peak bursts, • Full DBA techniques. The first approach is inefficient for satellite systems, as bandwidth is allocated in a way that does not take into account the real needs of a station; besides, the maximum source rate is usually unknown. As regards the full DBA techniques, these can exploit the channel capacity with good efficiency, since no capacity is reserved during inactive periods. Notwithstanding, the capacity request signaling channel may become overloaded during transient changes in traffic, leading to higher delays and congestion. Consequently, a mixed approach seems to be the most flexible choice, where each terminal is assigned some fixed channels of moderate capacity, while a number of DBA channels are used during peak traffic periods. As far as adaptive schemes are concerned, one of the challenging problems that engineers are called to grapple with is the implementation of these techniques in a GEO satellite system. The main problem stems from the high delay between the time instant that a request is sent to the NCC and the time instant at which the satellite terminal is informed about the bandwidth that has been allocated to it. This latency prevents immediate changes to the allocated capacity. Since a low latency entails better performance, a GEO satellite system represents the worst case (approximately 500 ms when the NCC is terrestrial-based or 250 ms when the majority of processing is supported by the satellite as a part of its on-board capability). Adaptive DBA schemes are generally categorized as either reactive or proactive algorithms. Reactive schemes take into account the current queue length, the packet loss and the average delay in order to react to traffic fluctuations, without trying to anticipate them. Compared to proactive al- gorithms, reactive algorithms are easier to implement and can utilize the channel capacity more efficiently. However, QoS requirements are not easily met, since the requests are delayed by sending them to the NCC, and do not therefore necessarily represent the current bandwidth needs. In [6], the authors proposed a novel predictive bandwidth allocation and de-allocation scheme, which frees up bandwidth allocated to connections that are unlikely to be used. The look-ahead horizon of k cells is introduced, where k =2.The scheme provides the lowest Call Dropping Probability (CDP) for real-time connections with respect to previous schemes. Even though reactive schemes may perform well in LEO satellite networks, they are not well suited to GEO systems owing to the high propagation delay. proactive schemes aim at analyzing the traffic and predicting the required bandwidth. Usually, this is realized by providing a predictor with data up Chapter 7: DYNAMIC BANDWIDTH ALLOCATION 211 to time t (e.g., with the queue lengths, the input flows and output flows); such data are used to make a prediction at time t of the aggregated traffic in the interval [t, t + k] (e.g., the traffic within the next superframe, where a superframe is the aggregation of k consecutive frames). Depending on the number of simultaneous traffic flows (i.e., TCP connections, application data streams) and the QoS model in use (i.e., DiffServ or IntServ), different traffic prediction techniques can be adopted. In a single-user per satellite terminal scenario, an IntServ-based QoS model will be assumed, whereas for a large aggregate of users per terminal, a DiffServ model seems more appropriate. When the number of data flows is very small, e.g., for a single-user per satellite terminal, traffic predictors may exploit the possibly known traffic patterns, like the TCP slow-start and the IntServ traffic information, in order to reserve the appropriate resources. If this is not viable, as in a DiffServ model approach, the traffic predictions can resort to utilizing the statistical properties of IP traffic. Hence, the required bandwidth can be estimated. In order to make adaptive predictions, i.e., capable of following changes in the traffic characteristics over time, the parameters of the predictor can be regularly updated. The performance of these schemes heavily relies upon the accurate prediction of future traffic. 7.2 DBA schemes for DVB-RCS scenarios In a DVB-RCS return link, users are multiplexed by means of a Multi Fre- quency - Time Division Multiple Access (MF-TDMA) scheme. The DVB-RCS standard [7],[8] permits full flexibility in the way the bandwidth is divided (see a feasible example in Figure 7.1, left upper corner [9]). The adopted solution in this Section consists of an independent division of both time and frequency axis, that is, bandwidth is divided into several carriers and the time duration of the superframe is divided into timeslots. Carriers do not have necessarily the same transmission bandwidth (different types of carriers are possible) and, at the same time, the timeslot duration can be different from one carrier to another. Return Channel Satellite Terminals (RCSTs) ask for some amount of system capacity to the NCC through capacity requests. In the DVB-RCS standard, three types of capacity request, from highest to lowest priority, are considered: CRA, RBDC, and VBDC. Free Capacity Assignment (FCA) usually is not taken into consideration by DBA schemes, since it may be granted by the NCC, but not requested [10]. Please refer to Chapter 1 for more details on these resource allocation methods. Note that the requests generated by all RCSTs in a beam constitute the inputs of the bandwidth allocation problem and in principle it is not necessary to consider how RCSTs generate requests. For each bandwidth allocation update (which is done on a superframe basis) the NCC sends a Terminal Burst Time Plan (TBTP) to the RCSTs. This message indicates the time 212 Tommaso Pecorella, Giada Mennuti Fig. 7.1: System model. See reference [9]. Copyright c 2006 IEEE. and the frequency that each RCST should use to transmit (see Figure 7.1 [9]). In any case, the standard does not give strict constraints on the algorithms to be used in the resource allocation process; hence, it is possible to develop advanced techniques by using the standard request types. The only weakness of the standard is related to the lack of information contained in the requests; hence, two requests of the same type will have to be considered as equal, even if the requesting RCSTs have to deliver different kinds of traffic (e.g., volume-based requests for high priority and low priority traffic). The next improvements in DVB-RCS-based allocation strategies will be focused on two topics, both related to a cross-layer approach. The first one will be to consider the effects of fading countermeasures; the second one will be to define a simple interface for upper layers, in order to develop a cross-layer QoS manager, able to tune the allocation process to the actual QoS requirements, possibly considering a pricing system, i.e., taking into account the user willingness to pay. A possible protocol architecture to support cross-layer interactions is proposed in sub-Section 1.6.2 referring to the BSM standard. Chapter 7: DYNAMIC BANDWIDTH ALLOCATION 213 7.3 Recent developments on DBA techniques 7.3.1 DVB-RCS dynamic channel allocation using control-theoretic approaches One of the main issues with proactive DBA is the accurate prediction of future traffic. Traffic predictors are usually affected by errors due to unex- pected network behaviors (e.g., packet loss, network congestion, etc.), TCP behavior and, more generally, uncertainty in the user interactions. Coupling the traffic predictors with appropriate control-theoretic techniques, however, allows maintaining the required QoS with an acceptable computational effort. In a DVB-RCS GEO satellite system, the NCC receives the bandwidth requests of each RCST and decides whether to satisfy or not these requests on the basis of a fair policy of resource sharing among all the RCSTs. In order to meet the desired QoS, both the request algorithm and the NCC allocation strategy are of paramount importance. In [11]-[13] the authors compared some different allocation strategies based on traffic prediction, assuming that each RCST is used to transmit a heavy aggregate of traffic. Figure 7.2 shows the proposed system model. It can be observed that the bandwidth controller must take into account the traffic predictions, the actual queue sizes and the packet scheduler behavior, to satisfy the bandwidth requests. In the figure, the NCC is depicted as a simple delay with a “disturb”, due to the possibility of denying a bandwidth request. Fig. 7.2: RCST system model. . another, either by incorporating Inter -Satellite Links (ISLs) or a ground-based hub station in each footprint. All these issues contribute in making DBA an essential approach for providing the proper. downlink spot-beams, the channel allocation will have to take into account also this aspect in order to maintain the overall QoS; this is particularly important in satellite- based switching systems. S. Marano, “Controlled Load Service Management in Int-Serv Satellite Access Networks , in Proc. of the Canadian Conference on Electrical and Computer Engineering 2004, Vol. 4, pp. 2193-2196, May