Jitter performance for QoS in Ethernet Passive Optical Networks Abhishek Dixit, Goutam Das, Bart Lannoo, Didier Colle, Mario Pickavet, Piet Demeester Department of Information Technology, Ghent University – IBBT, B-9050 Gent, Belgium abhishek.dixit@intec.ugent.be Abstract: We propose a new online dynamic bandwidth allocation algorithm for QoS in EPON, which provides a constant delay and jitter performance to higher and medium priority traffic classes while maintaining a high throughput. OCIS codes: (060.4250) Networks; (060.4252) Networks, broadcast 1. Introduction Emerging applications like voice-over-IP (VOIP), IPTV, broadband data, and peer to peer sharing require a guaranteed bound on many parameters like: bandwidth, packet delay (latency), delay variation (jitter), and packet- loss ratio. Quality of Service (QoS) refers to a network’s ability to provide bounds on some or all of the above parameters. The paper focuses on QoS issues in EPON networks. To achieve an effective resource allocation in EPONs, several schemes using dynamic bandwidth allocation algorithms (DBA) have been proposed, such as Interleaved Polling with Adaptive Cycle Time (IPACT) [1]. IPACT achieves high throughput but the variable cycle time of the algorithm is not suitable for the high priority jitter sensitive applications [2, 3]. To provide constant jitter performance, several fixed cycle time scheduling algorithms (such as Hybrid Slot-Size/Rate protocol (HSSR) [4], Cyclic-Polling-based Bandwidth Allocation with SLAs (CPBA-SLA) [3]) or approaches with separate cycle for each traffic class (such as Hybrid Grant Protocol (HGP) protocol [2]) have been proposed. The fixed frame (cycle time) algorithms limit the channel utilization in the context of highly bursty traffic. In HGP protocol, there is a separate cycle for each traffic class and there are idle periods between each cycle which limits the throughput. Thus, there is a need for an algorithm which provides constant delay variation to high priority traffic with high throughput. Also, with emergence of many applications (like interactive or streaming video); a service differentiation is needed for even medium priority traffic. An algorithm must be able to serve applications according to the specified parameter bound (leading to the parameterized QoS control). We propose the Delay-Aware Window Sizing (DAWS) approach (based on IPACT) to reduce the delay and the delay variation for high priority traffic. We achieve an improved jitter performance and a considerable higher throughput. For medium priority traffic, we propose the Delay-Aware Grant Sizing (DAGS) approach which helps to maintain the average delay according to the specified parameter and minimizes jitter. The simulation results show the effectiveness of our proposed algorithm. The rest of the paper is organized as follows. Section 2 presents our new protocol. Detailed performance analysis is done in Section 3. Section 4 concludes the paper. 2. Algorithm Based on the required bound, we categorize the traffic into three different classes: a) EF (Expedited Forwarding) – highest priority, delay sensitive traffic with constant bit rate such as voice transmission, b) AF (Assured Forwarding) – medium priority, delay sensitive traffic with variable bit rate such as video transmission and c) BE (Best Effort) – low priority traffic for non-real time data transfer such as e-mail applications. We will now discuss how low delay and jitter bounds are met for the EF and AF traffic class. DAWS (for EF traffic): For non-bursty EF traffic, the delay variation of the first departed EF packet between two consecutive transmission windows (inter-window jitter), maps the distribution property of the total EF delay sequence of an ONU [2]. EF packets with more fluctuation in their inter-window jitter tend to be over delayed or under delayed with respect to their mean value, and so the total EF delay sequence appears to be more dispersed. The inter-window jitter between the i th window and the (i+1) th window J i is given by (1), where D i is the delay of the first departed packet within the i th window [3].       (1) For predictable EF traffic, the grant-before-request (GBR) method has been suggested in [2]. In the GBR method, the EF traffic will be granted before a request is made from an ONU and the maximum delay bound of the