| |||||||||||||
Open Source Cluster File System Hits 1.0
First Production Version of the Lustre ™ File System Released The Lustre™ File System v1.0 for Linux 2.4.x has been released and is now available for free download. The Lustre™ File System is a scalable cluster file system that has been designed by CFS from the ground up to meet the demands of the world's largest high-performance compute clusters. Our state of the art object-based storage architecture redefines scalability and provides groundbreaking I/O and metadata throughput. This release of Lustre scales to thousands of nodes and hundreds of terabytes of storage-- with active development to support tens of thousands of nodes, trillions of files, and petabytes of data. Lustre is Open Source software developed and maintained by CFS under the GNU General Public License. Release notes and download information can be found at http://www.clusterfs.com/download.html . Information on Lustre Support, Training and Engineering services provided by CFS can be found at http://www.clusterfs.com/services.html . Customer profiles and experiences with Lustre deployments can be found at http://www.clusterfs.com/customers.html . The Open Source project homepage for Lustre can be found at http://www.lustre.org/ . December 15, 2003 (Log in to post comments)
Open Source Cluster File System Hits 1.0 Posted Dec 16, 2003 19:22 UTC (Tue) by TimCunningham (guest, #10316) [Link] Heh, fantastic. The download link that they send you in an e-mail has a rather broken SSL cert.. Mozilla complains twice, then asks me for a username and password. If I hit cancel it shows me the download page, and then they insist that if I link to this download, that I like to the registration page (so they can harvest more e-mail addresses)...For anyone else who wants to check this out, here are the download links directly... http://www.clusterfs.com/v1.0/lustre-1.0.tar.gz
Open Source Cluster File System Hits 1.0 Posted Dec 16, 2003 20:17 UTC (Tue) by BrucePerens (subscriber, #2510) [Link] I got to talk with one of their people, they had already realized there was a problem and will fix it.Bruce
Open Source Cluster File System Hits 1.0 Posted Dec 16, 2003 20:22 UTC (Tue) by proski (subscriber, #104) [Link] Did you notice that they call GPL "Software License and Terms of Use"? It looks like they treat GPL as a contract, not just as a license. That's why they make it so hard to get the software. They want to make sure that nobody gets the software without agreeing with their "Terms of Use". Understanding GPL is not required, but clicking "I agree" is.Your direct links circumvent their scheme. I wonder if they can use DMCA against you :-)
Open Source Cluster File System Hits 1.0 Posted Dec 16, 2003 20:27 UTC (Tue) by Ross (subscriber, #4065) [Link] As we all know, no terms of use are mentioned in the GPL. I think they arejust used to boilerplate "restrict all actions" types of agreements. I don't think it's anything more sinister. Someone should just send them a friendly note explaining the difference.
Open Source Cluster File System Hits 1.0 Posted Dec 16, 2003 20:37 UTC (Tue) by snitm (subscriber, #4031) [Link] It gets worse, if you go to http://www.clusterfs.com/download.html and highlight the text of the GPL; you can delete it and type anything you want in and then hit "Continue". So nobody _really_ has to agree to anything.
Open Source Cluster File System Hits 1.0 Posted Dec 16, 2003 21:57 UTC (Tue) by christopherbeard (guest, #17777) [Link] Thank you for your post. I hope that we have addressed all the issues that you have raised, and offer our apologies for any technical difficulties you may have had in accessing the code. Cluster File Systems, Inc. and the Lustre project have always been and continue to be strong supporters and advocates of Open Source software and we are very proud and pleased to offer up the 1.0 release of Lustre as a highly scalable and high-performance cluster file system for Linux. We invite you to stay up to date with product news, updates and software releases for Lustre at http://www.clusterfs.com/ lustre.html.
Open Source Cluster File System Hits 1.0 Posted Dec 17, 2003 16:28 UTC (Wed) by TimCunningham (guest, #10316) [Link] Thanks for the reply. Looks like it works now.
Open Source Cluster File System Hits 1.0 Posted Dec 16, 2003 19:35 UTC (Tue) by hjweth (guest, #1365) [Link] Is there a link to the man page for clusterfsck?
Open Source Cluster File System Hits 1.0 Posted Dec 16, 2003 20:11 UTC (Tue) by proski (subscriber, #104) [Link] They don't even have such utility. Quote from doc/addition-removal.lyx (from lustre-1.0.tar.gz):I object to any proposal that walks the MDS file system because it is not scalable in the face of trillions of files on the MDS (in particular I consider the fsck tool a useful debugging utility but not a tool useful for production). There are better solutions.
Open Source Cluster File System Hits 1.0 Posted Dec 16, 2003 20:54 UTC (Tue) by dsime (guest, #5764) [Link] I doubt that that has any relevance to the question/comment.
Open Source Cluster File System Hits 1.0 Posted Dec 17, 2003 0:16 UTC (Wed) by xorbe (guest, #3165) [Link] You missed the joke AND the question/answer!Q: man page for clusterfsck? (ha!) A: fsck isn't a supported utility. Jokes aside, how DO they recover then?
Open Source Cluster File System Hits 1.0 Posted Dec 17, 2003 9:35 UTC (Wed) by proski (subscriber, #104) [Link] As I understand it, MDS (metadata) is replicated across the cluster, so it should survive failures of single nodes. The data proper is kept on the nodes in ext2 or ext3 format.
Recovery Posted Dec 17, 2003 10:20 UTC (Wed) by pschwan (guest, #7699) [Link] > Jokes aside, how DO they recover then?When a Lustre server node fails in a non-destructive way -- such as a power outage, node crash, or network disconnection -- then it will restart in "recovery". During this recovery window, the metadata server (MDS) and object storage servers (OSS) execute a protocol which re-syncs the metadata. This protocol is not completely unlike what happens in a journalling file system, although it is somewhat more complicated. An example of what could happen is the following: When a user removes a file, the filename is first unlinked from the MDS; then the data object(s) for this file are deleted from one or more OSSs. Imagine the case of the unlink succeeding on the MDS, but before the OSS finishes it loses power. When the node reboots, there is data on an OSS which no longer has a filename associated with it on the MDS. These so-called "orphan" objects are cleaned up by the recovery protocol automatically, without the need for a fsck. There is one important case where a fsck can be required: after a destructive hardware failure. If an entire raid set fails, or if an MDS or OSS is restored from a tape backup, a fsck tool is needed to re-sync the servers. This tool is being tested now, and will be provided in a later Lustre 1.x release.
Open Source Cluster File System Hits 1.0 Posted Dec 16, 2003 21:08 UTC (Tue) by wcooley (subscriber, #1233) [Link] There's a joke here you've missed...
This is very good news Posted Dec 16, 2003 19:51 UTC (Tue) by emkey (guest, #144) [Link] The one major opensource component that has been lacking to make Linux a fully functional platform for running very large parallel jobs is now in place.
System V code? Posted Dec 16, 2003 23:37 UTC (Tue) by bojan (subscriber, #14302) [Link] Now, this looks to me like enterprise class stuff, therefore it must be stolen from System V. I'm calling Darl right now... ;-)
|
|||||||||||||