XtreemOS 2.0 released [LWN.net]

XtreemOS 2.0 has been announced; it is a grid-oriented distribution developed with support from the European Union. One of the core components appears to be the XtreemFS distributed filesystem. "XtreemOS brings new capabilties to Grids, such as easing job submission and monitoring, while providing a comprehensive security implementation and virtual organization management."

to post comments

XtreemOS 2.0 released

Posted Nov 17, 2009 18:41 UTC (Tue) by xlab (guest, #59190) [Link] (2 responses)

Some more info here:
http://tech.xlab.si/?p=913

xtreemFS

Posted Nov 18, 2009 4:19 UTC (Wed) by Alterego (guest, #55989) [Link] (1 responses)

I hope to read an LWN article about xtremFS , glusterFS and other distributed files sytems, wrt to fault tolerance or performances, and easyness of configuration (not like AFS !)

xtreemFS

Posted Nov 20, 2009 4:31 UTC (Fri) by jd (guest, #26381) [Link]

Well, some of the premiere experts on Linux filesystems write for LWN from time to time, so there's hope yet. In terms of clustered filesystems, I'd hope to see a reasonable range reviewed and also to see how some of the interesting features work in practice and how development approaches affect the product.

Example 1: NFSv4 supports RDMA, but how much does this buy you when the logic for the filesystem is in the kernel?

Example 2: Lustre uses VMS-style locking, whereas most other clustered filesystems use Unix/POSIX conventions. Forgetting the idea that one is good and the other is bad, WHEN is each method the right one to use and what limitations does that place on Lustre?

Example 3: P9000, the Linux port of the Plan9 clustered file system, is a fascinating concept but I've heard a lot of nasty comments about Plan9 code and architecture. Does P9000's innovative design get you results or grief?

Example 4: Polyserve's clustered FS is available for Linux but is certainly not open source. It's as commercial as they get. Commercial development can have high latency as commercial users don't update either kernels or add-ons routinely. For obvious reasons. I've heard that the Polyserve FS blows everything else away, but I've also heard about martians and the man in the moon. What's the reality? Are there filesystem or network features that it doesn't/can't use because of commercial considerations?

Afterthought: If Polyserve's developers are indeed producing the best network filesystem out there, is there anything we can beg/borrow/steal from their method of converting ideas into code to improve filesystem development in Linux in general?

Afterthought 2: There was a patch for Linux for Transactional TCP (T/TCP) which allows whole network transactions to be replayed or reversed. Is anyone looking at log-structured clustering file systems, if it is possible to use journals/logs with network filesystem activity? Or would that be the wrong level for such activity if you did want to add journalling?