学术文献库

Congestion control is vastly important in computer networks. Arising naturally from the bursty nature of Internet traffic, congestion plagues not only the network edge, but also the network core. Many remedies have been proposed to fight congestion; active queue management (AQM) is one such proposal. AQM seeks to prevent congestion by actively avoiding it. Some queuing disciplines such as Random Early Detection (RED) will prematurely drop a random packet (with some probability) when the queue nears capacity to signal the sender to back off. However, RED utilizes queue length as a mechanism to indicate congestion. On the other hand, the Controlled Delay (CoDel) queuing discipline uses queuing delay as an indication of congestion. The problem with both RED and CoDel are that they indiscriminately treat all packets the same. Normally implemented using a FIFO queue, CoDel simply enqueues and dequeues packets in a first-come, first-served manner. Priority queuing can be carefully utilized to selectively service packets utilizing the very same metric CoDel uses for AQM, queuing delay. That said, Least Slack Time First (LSTF), a multi-processor scheduling algorithm employs priority scheduling, which coincidentally, is also based on delay. In the context of computer networks LSTF can be applied in the control plane or in the data plane. At the control plane, LSTF functions across the entire network, but in doing so requires all intermediary routers to implement it; LSTF also requires support at the packet level in terms of a slack entry. Within the data plane, LSTF can be implemented as a queuing mechanism based on delay spent in the router (just like CoDel AQM). This paper applies data plane level LSTF to CoDel AQM to enable delay-based packet classification within the confines of the CoDel AQM algorithm.