A Hybrid EDF/FIFO Queue for Efficient Real Time Flow Handling S. Ould Cheikh El Mehdi, W. Fawaz, K. Chen L2TI Lab., institut Galil´ ee, University of Paris 13, 99, Av. J.B. Clement, 93 430 Villetaneuse, France saadbouh@l2ti.univ-paris13.fr, {fawaz, kchen}@galilee.univ-paris13.fr Abstract In an attempt to meet the stringent time constraints of real time flows, several scheduling algorithms have been proposed in the literature. Among such algorithms, the so-called EDF algorithm is widely known for its optimal management of flows with strict time constraints, when compared with the standard FIFO algorithm. But still, EDF is complex and expensive as far as implementation is concerned. As a main contribution in this paper, we therefore propose a hybrid EDF/FIFO queue management approach, which consists of a short EDF queue completed by a FIFO queue. This approach allows reducing EDF’s implementation complexity while making efficient use of its optimal flow management. Our simulation results underscore the main interest behind this proposal. Index Terms: - Real-time Scheduling, Quality of Service (QoS), Earliest Deadline First (EDF), Implementation optimization. 1. Introduction The Internet technology is becoming a de facto standard for almost all kind of communications, not only in wide area networks, but also in more restric- tive areas. Actually, more and more standard equip- ments and applications, as well as development man- power, are becoming IP-oriented. Internet has long been limited by the type of ser- vice provided to end users, which relies on the best effort concept basis. The network holds the sole promise of doing its best regarding packets’ delivery to their destination, and no more guarantees are of- fered to the end users’ traffic. This best effort type of service is quite suitable for the so-called elastic ap- plications (basically TCP traffics) which may toler- ate delay variations while compensating for eventual packet losses through retransmissions. In this regard, since the network offers a minimal service, ensuring better end to end service conditions is dealt with at the transport layer of end users’ applications. These applications were thus content with the minimal ser- vice context of the Internet. However, the perpetual development of the Inter- net is boosting the creation of new types of appli- cations, such as multimedia applications, and whose requirements can not be satisfied through the best effort service context. Indeed, these emerging appli- cations are presenting stringent real time constraints through the strict delay and/or bandwidth require- ments, which are needed by their generated traffic flows. The strictness degree of such requirements varies from one type of application to another. Meet- ing these various constraints implies the need to provide the different applications with differentiated Quality of Service (QoS) levels, which are adapted to their needs. In retrospect, several pioneer works have been con- ducted in the literature to deal with the QoS issues in packet routed multimedia networks. These activ- ities led particularly to the proposition of different packet scheduling techniques, among which one can cite Weighted Fair Queuing (WFQ) [1][2], and its variants [3]. Such scheduling techniques are gener- ally deployed within a multimedia network enforcing the IntServ or the DiffServ framework. The ratio- nale behind these scheduling techniques lies in guar- anteeing the bandwidth required by each data flow while bounding the response time of the data flow. Nonetheless, the response time resulting from the de- ployment of such scheduling techniques, referred to as share-driven scheduling algorithms, is function of both traffic burst size and the reserved bandwidth. This is especially true, since in the case of a bursty traffic, the end to end delay increases linearly with the increase of the maximal burst size. As a result, this may lead to the non-respect of real time pack- ets’ deadline. To cope with such a limitation, other scheduling techniques referred to as deadline driven scheduling algorithms are proposed for deployment in the network. In this article, our attention will be focused on one of such scheduling techniques, the so-called EDF algorithm (Earliest Deadline First) [4] [5][6][7], which is widely known in the context of real time traffic scheduling. The main idea behind EDF is the following: to 1-4244-0379-0/06/$20.00 ©2006 IEEE. 157