Improving Performance Bounds for Weighted Round-Robin Schedulers under Constrained Cross-Traffic

TL;DR

This paper refines theoretical performance bounds for weighted round-robin schedulers under constrained cross-traffic using network calculus, correcting previous incomplete proofs and demonstrating superior experimental results.

Contribution

It introduces resource-segregating policies, completes the proof that WRR schedulers belong to this class, and provides improved bounds with experimental validation.

Findings

Corrected delay bounds are slightly worse than previous estimates.

Experimental results outperform the state of the art.

Theoretical framework clarifies WRR scheduler classification.

Abstract

Weighted round robin (WRR) is an effective, yet particularly easy-to-implement packet scheduler. A slight modification in the implementation of WRR, interleaved weighted round robin, has been proposed as an enhancement of the initial version and has been recently investigated. Network calculus is a versatile framework to model and analyze such network schedulers. By means of this, one can derive theoretical upper bounds on network performance metrics, such as delay or backlog. In our previous work, we derive performance bounds by showing that both round-robin variants belong to a class called bandwidth-sharing policy; however, the proofs are incomplete and thus, we cannot conclude that the round-robin schedulers are bandwidth-sharing policies (under variable packet sizes).To that end, in the subsequent erratum, we introduce so-called resource-segregating policies and show the round-robin schedulers to be members of this class. We first present our original work, as published in [CNS22-1], and then the erratum correcting the previously mentioned shortcoming. In our erratum, we provide slightly worse delay bounds compared to [CNS22-1]; yet, across all our experiments, they significantly outperform the state of the art.