An effective scheduler for IP routers

At the Communications and Telematics Laboratory of the University of Coimbra is being developed a router prototype with the aim to provide QoS to different traffic classes. One of the most important mechanisms of this router is the IP packet scheduler. It is well known that the common scheduling discipline of current routers (first come first serve) makes them useless when QoS is needed-a different type of scheduler must be used. Our first idea to overcome this problem was to use a simple, open, and available scheduler, easy to adapt to the system we wanted to implement. We thought of the WFQ discipline, and, as we are using a testbed of Intel machines running FreeBSD OS, we admitted that the WFQ/ALTQ implementation would be an interesting choice. Nevertheless, a broad set of tests carried out at our laboratory proved the contrary. Most important, these tests guided us to a deep knowledge of the problems, and causes, that can weaken the effectiveness of IP schedulers. Given the importance of that surplus information, we decided to implement our own scheduler. The idea was to take advantage of a pragmatic view of scheduling activities to construct a scheduler with the best possible characteristics (but also very simple) able to reach very good performance levels. This paper presents the scheduler that resulted from our attempts. The proposed scheduler was subject to a set of tests that proved its ability to effectively differentiate traffic classes. The results of these tests are also presented and analyzed