Hi, I have tried Lustre 1.6 beta3 on a small configuration and I find it very easy to configure (it is much more simpler than previous configuration process). But I found a strong constraint : there must be only one MGS for all Lustre FS mounted on a client. I tried with multiple MGS (even if the documentation said it is forbidden ... ) and the second client mount request goes to the first MGS (the second MGS addr in the second mount command seems to be ignored). This limitation is an issue if we make cross site Lustre mounts (we are studying this on 10 Gige class WAN) or cross mount between on site clusters (if we start with a multiple MGS conf, we cannot cross mount later without rebuilding all lustre FS) Why such limit ? For exemple the rule should be that MDS/OSS must be attached to only one MGS but a client can mount from multple MGS (like today with MDS) JC
Hi JC - How interesting, and I think I agree, so let''s call it a bug and fix it. It''s probably a small issue in fact. - Peter -> -----Original Message----- > From: lustre-devel-bounces@clusterfs.com > [mailto:lustre-devel-bounces@clusterfs.com] On Behalf Of > Jacques-Charles Lafoucriere > Sent: Tuesday, May 30, 2006 2:39 PM > To: lustre-devel@clusterfs.com > Subject: [Lustre-devel] Lustre 1.6 and MGS > > Hi, > > I have tried Lustre 1.6 beta3 on a small configuration and I > find it very easy to configure (it is much more simpler than > previous configuration process). > But I found a strong constraint : there must be only one MGS > for all Lustre FS mounted on a client. > I tried with multiple MGS (even if the documentation said it > is forbidden ... ) and the second client mount request goes > to the first MGS (the second MGS addr in the second mount > command seems to be ignored). > > This limitation is an issue if we make cross site Lustre > mounts (we are studying this on 10 Gige class WAN) or cross > mount between on site clusters (if we start with a multiple > MGS conf, we cannot cross mount later without rebuilding all > lustre FS) > > Why such limit ? > For exemple the rule should be that MDS/OSS must be attached > to only one MGS but a client can mount from multple MGS (like > today with MDS) > > JC > > > _______________________________________________ > Lustre-devel mailing list > Lustre-devel@clusterfs.com > https://mail.clusterfs.com/mailman/listinfo/lustre-devel > >
Let''s call it a feature enhancement instead :) Bugzilla 10586 https://bugzilla.clusterfs.com/show_bug.cgi?id=10586 Peter J. Braam wrote:>Hi JC - > >How interesting, and I think I agree, so let''s call it a bug and fix it. >It''s probably a small issue in fact. > >- Peter - > > > >>-----Original Message----- >>From: lustre-devel-bounces@clusterfs.com >>[mailto:lustre-devel-bounces@clusterfs.com] On Behalf Of >>Jacques-Charles Lafoucriere >>Sent: Tuesday, May 30, 2006 2:39 PM >>To: lustre-devel@clusterfs.com >>Subject: [Lustre-devel] Lustre 1.6 and MGS >> >>Hi, >> >>I have tried Lustre 1.6 beta3 on a small configuration and I >>find it very easy to configure (it is much more simpler than >>previous configuration process). >>But I found a strong constraint : there must be only one MGS >>for all Lustre FS mounted on a client. >>I tried with multiple MGS (even if the documentation said it >>is forbidden ... ) and the second client mount request goes >>to the first MGS (the second MGS addr in the second mount >>command seems to be ignored). >> >>This limitation is an issue if we make cross site Lustre >>mounts (we are studying this on 10 Gige class WAN) or cross >>mount between on site clusters (if we start with a multiple >>MGS conf, we cannot cross mount later without rebuilding all >>lustre FS) >> >>Why such limit ? >>For exemple the rule should be that MDS/OSS must be attached >>to only one MGS but a client can mount from multple MGS (like >>today with MDS) >> >>JC >> >> >>_______________________________________________ >>Lustre-devel mailing list >>Lustre-devel@clusterfs.com >>https://mail.clusterfs.com/mailman/listinfo/lustre-devel >> >> >> >> > > >