Cluster File Systems and Sun to Integrate Lustre with OpenSolaris ZFS

Well, that was a nice start for the software townhall today and what a great proof point for the growing adoption of Solaris in the HPC community. You can read the full press release from Cluster File Systems that talks about how they will be using the OpenSolaris ZFS file system for Lustre Object Storage Servers and Lustre Meta Data Servers. Please do read the fine print, Cluster File Systems will not just be using ZFS for their Solaris implementation, they will be using ZFS as the basis for all versions of Lustre. Now that is cool. BTW, almost everyone has heard of ZFS, if you would like to understand a little more about how ZFS works, check out the ZFS source code tour on the OpenSolaris ZFS community.

Of course the main purpose of our software townhall for press, analysts, and bloggers, held this morning in San Francisco, was to talk about the future of Solaris and Project Indiana which is being built on OpenSolaris. Here are slides we presented. You can of course just go to the Project Indiana pages on OpenSolaris.org, but here are some of the highlights:

  • Project Indiana will develop an OpenSolaris binary distribution
  • The features of that distribution are being defined by the OpenSolaris community, but goals for the project include easier instllation, network-based package management, and ZFS as the default file system
  • We expect test releases to be available in Fall 2007 with the first full releases to follow in Spring 2008.
  • Sun will provide support to customers who have a Solaris support subscription and want to develop on, be an early adopter of, and use the latest Solaris innovations that are delivered through this OpenSolaris binary distribution. There will be a different set of SLAs for support of Project Indiana, while our traditional enterprise support will continue to be offered for Solaris. Sun's support policy of one Solaris subscription for every server that a customer wants Solaris support on will continue.

    There are a lot more details than I have time to cover, but take a look at the slides from today, and join the OpenSolaris community to find out more, and to help Sun make one of the world's leading operating systems even better.

    • Posted on: Jul 12, 2007

    Posted by: marchamilton

    Category: General

    Tags: none

    Permanent link to this entry

    Comments:

    I do wonder how this will work since it is well known that the licences are incompatible.
    Will they use a FUSE implementation? Or will they work somehow around that?
    And can others adopt the situation to run ZFS on Linux as well?
    Regards,
    liquidat

    Posted by liquidat on July 16, 2007 at 03:53 PM PDT #

    I don't believe Cluster FS plans to try to mix licenses or rely on a full Linux kernel level ZFS implementation. On Linux, they plan to have Lustre write to the DMU level and supply a dynamically linked libzpool that sits outside the Linux kernel.

    Posted by Marc Hamilton on July 16, 2007 at 04:26 PM PDT #

    See my initial link to ZFS Source Code Tour for more info on DMU and libzpool. http://www.opensolaris.org/os/community/zfs/source/

    Posted by Marc Hamilton on July 16, 2007 at 04:31 PM PDT #

    But Lustre and (by association) ZFS needs to be able to write through the Linux kernel in order to write out to the block devices with any kind of performance. Unless CFS will be using ZFS entirely in userspace there would need to be _some_ integration with the Linux kernel. CFS is all about High Performance; I'm fairly certain ZFS won't be running from userspace when CFS deploys it... Sun could've easily granted CFS a license to use ZFS natively in Linux. No?

    Posted by Mike Snitzer on July 16, 2007 at 05:24 PM PDT #

    Post a Comment:
    Comments are closed for this entry.

    This blog copyright 2009 by marchamilton