English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Packet buffers in datacenter switches are shared across all the switch portsin order to improve the overall throughput. The trend of shrinking buffer sizesin datacenter switches makes buffer sharing extremely challenging and acritical performance issue. Literature suggests that push-out buffer sharingalgorithms have significantly better performance guarantees compared todrop-tail algorithms. Unfortunately, switches are unable to benefit from thesealgorithms due to lack of support for push-out operations in hardware. Our keyobservation is that drop-tail buffers can emulate push-out buffers if thefuture packet arrivals are known ahead of time. This suggests that augmentingdrop-tail algorithms with predictions about the future arrivals has thepotential to significantly improve performance.  This paper is the first research attempt in this direction. We proposeCredence, a drop-tail buffer sharing algorithm augmented with machine-learnedpredictions. Credence can unlock the performance only attainable by push-outalgorithms so far. Its performance hinges on the accuracy of predictions.Specifically, Credence achieves near-optimal performance of the best knownpush-out algorithm LQD (Longest Queue Drop) with perfect predictions, butgracefully degrades to the performance of the simplest drop-tail algorithmComplete Sharing when the prediction error gets arbitrarily worse. Ourevaluations show that Credence improves throughput by $1.5$x compared totraditional approaches. In terms of flow completion times, we show thatCredence improves upon the state-of-the-art approaches by up to $95\%$ usingoff-the-shelf machine learning techniques that are also practical in today'shardware. We believe this work opens several interesting future workopportunities both in systems and theory that we discuss at the end of thispaper.

Credence: Augmenting Datacenter Switch Buffer Sharing with ML Predictions