Maintaining Consistency in a Massively Parallel Environment

Making changes to systems in a highly-distributed, massively parallel environment can result in system inconsistencies. Change requests can time out or result in error. It can also take hours for updates to get scheduled or complete.

This talk describes how Google is using its proprietary package manager to maintain consistency across its server farm. We describe some of the design decisions and features that allow us to do things traditional package managers cannot in order to ensure consistency. We also discuss creative ways projects are leveraging the architecture and our infrastructure to distribute changes reliably and consistently.

Dinah McNutt is a Release Engineer at Google, Inc. A graduate of MIT, she has been involved with systems administration since the mid-1980’s. Some of her accomplishments include writing the Daemons & Dragons column for UNIX Review</i> magazine and serving on the USENIX LISA program committee several times (including once as chair).

She has 20 years of commercial release engineering experience and has released all types of UNIX-based software from shrink-wrapped through Web-based services to appliances.