Mertol Ozyoney
2007-Nov-06 14:33 UTC
[Lustre-discuss] Copying data from one Lustre system to an other lustre system
Hi; Customer have 2 x lustre system. Two independen MDS clusters looking after their individual OSS''s. We need to copy 80 TB of data from one system to the other system. One full copy + 6 incremental copies per week. I plan on offering a few data mover servers. Can I use one IB HCA dedicated to source lustre and another HCA dedicated to the target lustre, so that systems qill not have to share bandwith through one HCA? By the way do anybody have any scripts ready to do the job above? regards <http://www.sun.com/> http://www.sun.com/emrkt/sigs/6g_top.gif Mertol Ozyoney Storage Practice - Sales Manager Sun Microsystems, TR Istanbul TR Phone +902123352200 Mobile +905339310752 Fax +902123352222 Email <mailto:Ayca.Yalcin at Sun.COM> mertol.ozyoney at Sun.COM -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.lustre.org/pipermail/lustre-discuss/attachments/20071106/5a4db9c2/attachment-0002.html -------------- next part -------------- A non-text attachment was scrubbed... Name: image001.gif Type: image/gif Size: 1257 bytes Desc: not available Url : http://lists.lustre.org/pipermail/lustre-discuss/attachments/20071106/5a4db9c2/attachment-0002.gif
Daire Byrne
2007-Nov-06 17:21 UTC
[Lustre-discuss] Copying data from one Lustre system to an other lustre system
Mertol, You probably want to check out the "e2scan" work that is being done atm. There''s a bugzilla entry for it (NERSC). As this was not available to us a couple of months ago we came up with a way to snapshot both MDSs and then run "rsync --itemize-changes" between them. This builds a filelist that we can then split into deletes and copies. For the copies you can split it N times and use multiple "rsync --files-from=" instance across many clients. The filelist generation takes around 6-8 hours and the copies around 5 hours for our 72Tb filesystem. Fast enough for us to get a backup through every 24 hours. But like I said a better/faster longterm solution is to use e2scan. Daire ----- "Mertol Ozyoney" <Mertol.Ozyoney at Sun.COM> wrote:> _______________________________________________ > Lustre-discuss mailing list > Lustre-discuss at clusterfs.com > https://mail.clusterfs.com/mailman/listinfo/lustre-discuss