Collaboration Summit 2014

As storage density continues to increase faster than bandwidth and access speed, servers become increasingly prone to media failure and take longer to repair, reducing availability. This issue is already severe and continues to worsen, while practical solutions have so far proved elusive. We present a new approach to verifying filesystem structural integrity incrementally and online, based on extending the allocation group concept used by Ext4, Tux3 and other file systems. Where inconsistency is detected, the necessary information for online repair is provided, in contrast to checksum methods, which fail to detect certain structural issues and provide inadequate guidance for repair. With these techniques, we anticipate a future where taking file systems off-line for repair becomes largely unnecessary.

The audience is anybody operating a server with high volume storage, and anybody who cares about availability of their storage. This will be of particular interest to data center operators with both high volume storage and high availability requirements.

A practical solution to the online filesystem checking and repair problem will lower data center costs, increase availability and lower risk of data loss. Personal workstation users will also benefit from removing the annoyance of periodic offline file system checking.