File Systems and Persistent Memory: Why We Won't Change as Fast as We Should [linux.conf.au 2014]

A lot of work is ongoing in making the Linux kernel IO stack fast enough to keep up with the current crop of high speed SSD devices. These devices can sustain hundreds of thousands of IO requests per second and our existing stack throttles this down substantially. To make matters even more challenging, we have a new wave of technology that promises persistent storage at roughly the same speed as today's DRAM which implies we need to be able to support millions of IO requests per second. This talk will give an overview of the ongoing work in the Linux kernel with a focus on the file system work. We will cover some of the proposed ways to use these parts, both in the traditional block oriented ways and as byte addressable storage. The talk concludes with a discussion of why these new interfaces will be less important than maintaining our traditional, if somewhat slower, block oriented interfaces and will share any performance data we have. Ric Wheeler Ric works at Red Hat as the senior manager of the kernel file and storage team. He spent ten years at EMC in the Symmetrix and Centera groups, four years at The Open Group's Research Institute and four years at Thinking Machines working on the CM5 operating system. Back in graduate school, he worked on the MOSIX process migration system. Ric is a frequent speaker at Linux Foundation events, past chair of the USENIX FAST conference and a member of the Linux Foundation's Technical Advisory Board. http://linux.conf.au/schedule/30182/v...