POHMELFS: A New But Old Distributed Linux File-System
POHMELFS is new in terms of being an old but completely re-designed distributed file-system. The new POHMELFS serves as a POSIX front-end to the Elliptics network, which has already been used in production cases for multiple years and handled billions of objects to one petabyte in storage.
The original POHMELFS is also the work of Evgeniy Polyakov and was first announced in early 2008. The distributed file-system made it into the Linux 2.6.30 kernel in the staging area, but now this just-published code is a brand new incarnation of the Parallel Optimized Host Message Exchange Layered File System that's now built around Elliptics (though it's actually named after похмелье, the Russian word for hangover). POHMELFS is designed to be a superior replacement to NFS.
From the release announcement to Linux kernel developers, "Main goal of this storage is ability to work with multiple datacenters out of the box (replication unit aka group fits quite well into single datacenter abstraction) and is aimed at high IO operations per second workload type instead of boring bulk MB/s IOs in 21 century. Getting its p2p nature bulk IO is not an issue either."
Additionally, "Pohmelfs is a POSIX frontend to elliptics. It supports weak synchronization between mounted nodes in that regard, that data read/written into local page cache is not synced with the storage until timeout fires. Directory content when read is populated directly into dentry/inode cache instead of doing that per-entry to make directory reading fast."
While the POHMELFS code is now published, there's still some items the developers wish to implement like Qurorum read support, an HTTP compatibility mode for this distributed file-system, column read-write, and to replace the existing POHMELFS code in the mainline tree with this brand new implementation.
More information on POHMELFS is available here along with information on Elliptics.
6 Comments
