The Atlas Cluster Trace Repository
George Amvrosiadis, Michael Kuchnik, Jun Woo Park, Chuck Cranor, Gregory R. Ganger, Elisabeth Moore, and Nathan DeBardeleben
Many researchers evaluating cluster management designs today rely primarily on a trace released by Google due to a scarcity of other sufficiently diverse data sources. We have gathered several longer and more varied traces and have shown that overreliance on the Google trace workload is leading researchers to overfit their results. We have created the Atlas cluster trace repository to aid researchers in avoiding this problem. This article explains the value of using and contributing to Atlas.
To access this content, please purchase a USENIX Membership.