Elastic weighted fair scheduling method and system based on hardware implementation

Abstract

The invention provides an elastic weighted fair scheduling method and system based on hardware implementation. The method comprises the following steps: judging whether scheduling objects in forwarding channels need to be transferred from a standby list to an active list according to the storage attributes of the scheduling objects corresponding to the channels and the data to-be-sent states of the forwarding channels before each polling scheduling; if so, obtaining supplementary particles of deficit counters according to the numerical values of the deficit counters corresponding to the forwarding channels; meanwhile, modifying the counting values of the deficit counters corresponding to the forwarding channels according to the weight coefficient ratios corresponding to the forwarding channels, the data to-be-sent states of the forwarding channels at present and the supplementary particles of the deficit counters at present; and otherwise, keeping the counting values of the deficit counters corresponding to the forwarding channels. According to the elastic weighted fair scheduling method and system provided by the invention, excessively added values of the deficit counters in a refreshing process are reduced to reduce the scheduling jitter and improve the scheduling precision.

Description

technical field [0001] The invention relates to the technical field of network communication, in particular to a hardware-based flexible weighted fair scheduling method and a scheduling system. Background technique [0002] With the development of IP services, especially the rapid growth of multimedia services such as audio and video, traditional IP networks must not only provide traditional types of data services, but also support new services with high bandwidth and delay requirements, that is, require A traditional IP network distinguishes different services and provides certain QoS guarantees. The full English spelling of QoS is: Quality of Service, translated as service quality. [0003] The DiffServ model in the QoS model defined by the IETF generally implements most of the complex QoS functions through edge nodes, such as classification, traffic regulation, marking, etc., and the intermediate nodes perform different PHB on the traffic according to the differential se...

AI Technical Summary

Problems solved by technology

[0008] In the traditional MDRR algorithm, before scheduling, it is necessary to specify a fixed supplementary granule. The fixed supplementary granule is represented by granularity and cannot be changed. During each scheduling process, when switching scheduling rounds, if the After the deficit counter is corrected, the scheduling node corresponding to the switched deficit counter will still be scheduled and owe more values; considering that the counter implemented by digital logic has a bit width limit, after this event occurs many times, the deficit counter will occur Flip to become a huge positive number, making the final scheduling ratio inaccurate; so, in order to avoid the above phenomenon, the value of the supplementary particle is usually set to the maximum length of the message in the network transmission process, called MTU, that is, to set the granularity = MTU; however, in many cases, the data packets are not large. If the debt counter is filled according to the weight multiple of the MTU each time, it will inevitably lead to fewer rounds of switching when the traffic is small packets; this will As a result, the dispatched traffic is not smooth enough, there are sudden traffic situations, and the final accuracy also fails to meet user needs

Images