Vault 2017

Providing good performance for directory operations in a scale-out storage solution like gluster can be challenging. Gluster relies on a framework called Distributed Hash Table (DHT) for hashing and distributing files, which makes it necessary to aggregate contents of a logical gluster directory from multiple physical directories on the underlying constituent file systems, known as bricks, for operations like readdir.

In this talk, Manoj Pillai will present an extensive evaluation of the performance of gluster directory operations for real-life scenarios. Raghavendra Gowdappa and Poornima Gurusiddaiah will then describe the following approaches for improving directory performance, and assess their effectiveness: readdir-ahead (prefetching), parallel readdir, readdir-optimize (filter duplicate directories that result from gluster's directory creation approach) and directory readdirplus.