An Adaptive Resource Assignment Architecture for the EuroSkyWay system using a Neural Network Burstiness Predictor

Abstract: The efficiency oJ use oJ the resources is an important task in the operation oJ a data communication system. An optimal way is to apply resource management policies based on dynamic scheme other than the fvced ones without penalising the end user. Satellite systems that, even if supporting static Bandwidth Allocation, provides mechanisms Jor a dynamic solution to the Resource Management with respeet to Real Time Variable Bit Rate and non-real Time Variable Bit Rate traffie sources. In order to meet the expectations oJ network users, in terms oJ parameters such as cell loss rate and call bloeking probability, the use oJ network resourees must be careJully eontrolled On the other hand, resources should be handled so that the satellite network operator revenue is maximised over the time. An advaneed traffic management architecture based on the Dynamic Resouree Assignment allows a broadband satellite system sueh as EuroSkyWay (ESW) to adapt dynamieal/y the resources oJ provided conneetions to diversified traffie supported by allowing to ehange dynamically the resources oJ each connection. Eaeh source is characterised by its Traffie Descriptor which define traffic source in terms oJ emission Jeature sueh as Peak Cell Rate, Sustainable Cell Rate and Maximum Burst Size. Traffic Deseriptors statically established beJore the connection set-up are part oj the Traffie Contraet. When the eonneetion is admitted a Traffie Contraet is defined as a commitment from both user side and network side. The mechanisms oJ the ESW dynamic assignment aims to exploit variations oJ Burstiness exhibited by the real time and non real time Variable Bit Rate traffic sources to perform an optimised resourees redistribution oJ those resources to other eonneetions (pre-emption). Although the ESW dynamic assignment mechanism is a simple way to assign/deassign resources to eonnections, the efficiency oJ it depends on the accuracy oJ the traffic source description; inaccurate assessment oJ the arrivaI process shall cause an overhead and a degraded utilisation oJ unused resourees within the system. Moreover, mainly due to the round trip delay needed to perform a complete handshake between terminaI and the on-board processor, a degradation oJ the CDValso oeeurs as a eonsequence oJ a frequent usage oJ Resouree Request Mechanisms. The number oJJrames assigned to the traffie souree should be strictly related to the preventive assessment oJ the activity period that the bursty souree wil/ exhibits in thefuture. In this paper, an approachJor predictingJuture behaviours oJthe traffic by means oJ a Neural Network Burstiness Predictor arehitecture Jor the EuroSkyWay system is presented The approach is ah/e to compute the expeeted resource request in real-time implementing direct adaptive traffic resouree assignment without an analytical representation oJ the traffic, but using a sub-symbolic representation oJ it. Moreover, the distributed and parallel architecture of NNs can be exploited in hardware implementations, largely available on market, which provide short and predictable response times.