A weighted fair queuing finishing tag computation architecture and implementation

Abstract

This paper investigates a custornimd systems, work conserving schedulm will be examined. The ideal scheduling discipline is Generalised Processor Sharing (~ps). Packets are organised into separate logical queues with each nonempty queue visited in and an inffitesimally fair allocation of bandwidth and offers protection, but it cannot h'ansmit packets as entities. Round Robin is the simplest emulation of GPS and serves packets from each nonempty queue while WRR provides connection based bandwidth guarantees. Both approximate GPS reasonably well . when all packets have the same size. Deficit Round Robin can handle variable size packets and is suitable when packets are irnplemcatation for XiUm Virtex Pro board of a Weighted Fair Queuiq gvFQ) tag scheduler that bas tbe capacity to *ewe 8,000 i u d k i d d scpsionr at over lOOMHz The implementation is sctndly upsble of b a n d l i up to 64,OOO different sessions an excellent hardware cosl for a next generation terabit router. r e q n i~g the size FPGA. Thb represene small amount of d a b served *om each queue. GPS achieves

QGWFQS: A Queue-Group-Based Weight Fair Queueing Scheduler on FPGA

“…McKillen [6] proposed an FPGA implementation of the WFQ scheduling algorithm, which utilizes a pipeline approach to calculate the finish tag for each incoming packet, even if they originate from the same queue. However, this computational approach introduces redundancy requirements on the maximum sorting capacity supported by the sorting circuit, as it transmits the tags of sequentially arriving packets from the same queue to the sorting circuit.…”

“…Therefore, accurately computing v(t) in hardware requires obtaining precise values of C(t) at any given moment, which is challenging in FPGA due to the discrete nature of time and transmitted bits. The packet tag calculation method, proposed by McKillen [6], based on WFQ assumes that C(t) is constant, which can adversely affect the accuracy of v(t), S(p j i ), and F(p j i ) calculations when C(t) changes, thereby impacting scheduling fairness.…”

“…The resource utilization of the Tag Calculator circuit is presented in Table 1, alongside the resource utilization of the method proposed by McKillen [6], which serves as a basis for comparison. Our approach supports tag computation for 16 k queues across 512 queue groups, while the comparative work needs to be instantiated 512 times to achieve a similar effect.…”

Estado Del Arte De Scheduling Algorithms (Algoritmos De Programación De Envío)

“…Por otra parte se tienen los trabajos realizados en [39], [40], [41], [42] y [43] en los cuales se aplica a situaciones, el concepto de IWFQ.…”