Hello All, I have read some papers about lustre and now I am dipping into the source code. I have some questions about lustre. Lustre is good at fault tolerant for fail over(in the manul''s term), but not refer to fail out failure, for example when the OST is fail stop or the medium is damaged. GPFS has supportted quorum and Google File System used three replicas for data availability, So I think Lustre should also resolve this problem. It is important for storage service. I want to ask if lustre Realy has mechanism to resolve this problem or some future plans to solve it? I would be appreciated for any reply! Thanks. ----------------------- wuguangjun at software.ict.ac.cn guangjunMSN at hotmail.com -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mail.clusterfs.com/pipermail/lustre-discuss/attachments/20070917/7687d741/attachment.html
Pravin Shelar
2007-Sep-18 15:34 UTC
[Lustre-discuss] [Lustre-devel] help to lustre fail out failure?
On 9/17/07, guangjun.wu <wuguangjun at software.ict.ac.cn> wrote:> > > > > Hello All, > > I have read some papers about lustre and now I am dipping into the source > code. I have some questions about lustre. > > Lustre is good at fault tolerant for fail over(in the manul''s term), but not > refer to fail out failure, for example when the OST is fail stop or the > medium is damaged. > > GPFS has supportted quorum and Google File System used three replicas for > data availability, So I think Lustre should also resolve this problem. It is > important for storage service. > > I want to ask if lustre Realy has mechanism to resolve this problem or some > future plans to solve it?AFAIK, to handle this type of failure, multi-server (OST) striping ( raid 1, 5 or 6 ) is required which will be supported in Lustre 2.0 onward. Current Lustre depend on (multi-port) storage device with high availability features. You can use DRDB* setup for all OSTs if u want software only solution to have some high availability. But I not sure this type of setup is supported by cfs or not. --pravin. [*] Distributed Storage which is just introduced to Linux kernel is promising alternative to DRBD, not sure how stable it is.> > I would be appreciated for any reply! Thanks. > > ----------------------- > > wuguangjun at software.ict.ac.cn > > guangjunMSN at hotmail.com > > > _______________________________________________ > Lustre-devel mailing list > Lustre-devel at clusterfs.com > https://mail.clusterfs.com/mailman/listinfo/lustre-devel > >