Parallel Platters: File Systems for HPC Clusters Part Three

As clusters require faster and faster access to data, while maintaining reliability, new filesystems are being developed to try to break the bottlenecks and provide scalable solutions. The next generation of filesystems, dubbed object based file systems, help break performance barriers by allowing data devices to manage where data is stored.

The first part of this series covered distributed file systems, with a focus on those using NFS as the underlying protocol. In the second part, we examined three parallel file systems based on more traditional architectures: GPFS from IBM, Ibrix, and MPFSi from EMC.

In this third and final installment, I will cover what might be described as the next generation of parallel file systems, object based file systems. As with the previous installments, I won’t be able to cover all of the file systems which fit the description due to space considerations, but will try to focus…

1. Saving Yourself with Data Replication
2. One Billion Dollars! Wait... I Mean One Billion Files!!!
3. Bcache Testing: Large Files and a Wrap-Up
4. Storage Monitoring via SystemTap
5. Bcache Testing: Metadata