Dan Luu 9/30/2015

This article explores the impact of 'limping' hardware—components that are degraded but not fully failed—on distributed systems like HDFS, Hadoop, and Cassandra. Citing research by Thanh Do et al., it explains how a single slow node can be more detrimental than a dead one, causing dramatic performance drops across entire clusters, and discusses the lack of robustness in common systems against such failures.

Slowlock

Comments

No comments yet

Be the first to share your thoughts!

Browser Extension

Get instant access to AllDevBlogs from your browser

Top of the Week

No top articles yet