Cover Image
QR Code

The Tail at Scale.

The article discusses high-latency episodes in large online services and tail-tolerant software techniques that form a predictable whole out of less-predictable parts to reduce the episodes' effect on whole-system performance. The article notes variability in latency distribution or response time to...

Full description

Published in:Communications of the ACM Vol. 56; no. 2; pp. 74 - 81
Main Authors: DEAN, JEFFREY, BARROSO, LUIZ ANDRÉ
Format: Article
Published: Association for Computing Machinery Feb2013
Subjects:
Web services
Distributed computing software
Fault-tolerant computing
Search engine optimization
Google (Web resource)
Computer performance
Computer software
Online Access:View this record in EBSCOhost
Summary:The article discusses high-latency episodes in large online services and tail-tolerant software techniques that form a predictable whole out of less-predictable parts to reduce the episodes' effect on whole-system performance. The article notes variability in latency distribution or response time to user actions can be reduced by parallelization of sub-operations across different machines. Topics include system utilization that avoids overprovisioning, performance variability, and the Google search engine.