News from the world of filesystems and storage devices
The annual USENIX Conference on File and Storage Technologies (FAST) is one of the premier events for researchers and practictioners working in the fields of filesystems, storage devices, and related technology. This year's conference, FAST '11, has just concluded, and I've been catching up on the various information that was shared at the conference. This is an information-packed event and every paper is worth reading. Often, trying to keep up with events like these can be overwhelming, so let me offer two "guides" to help you as you embark upon your study:
- At StorageMojo, Robin Harris has a nice summary post: Great Work at FAST '11. As Harris observes, a lot of the research attention nowadays continues to be devoted to flash storage devices and how to use them effectively. From what I've read, this year's transition from 34 nanometer flash to 25 nanometer flash is not the slam dunk you might think it would be.
- At CloudFS, Jeff Darcy looks at the conference from the perspective of Cloud Computing in his post: Checking out the Competition. Darcy's post has some great references to a number of cloud filesystem projects, including OrangeFS and Ceph.
As for myself,
- I liked the GIGA+ paper that was highlighted by both Harris and Darcy: Scale and Concurrency of GIGA+: File System Directories with Millions of Files
- and I also enjoyed the NetApp paper on proximal I/O: Improving Throughput for Small Disk Requests with Proximal I/O. I appreciated the practical aspects of this paper, as the authors spent a fair amount of time exploring the various tradeoffs that are possible in filesystem design, and explaining why it's hard to craft a single design that works for all situations.
At Perforce, we are quite concerned with filesystem implementation details. Perforce customers make extreme demands on their file systems: Perforce servers that manage terabytes of versioned data are common, and petabyte installations are no longer unique. Our performance team spends a lot of effort studying the behaviors of different file systems under different workloads in order to keep up to date with the best available current implementations. In the world of version control, system sizes grow steadily; before you know it, you have decades of historical data under management! As your installation grows and your needs increase, please feel free to contact us to discuss your particular situation and we'll be pleased to work with you to get the best possible system in place.