Hi You should do this comparison with ldiskfs (a disk file system included with lustre, just ext3 with serious fixes). ldiskfs has the advanced features like extents, buddy maps etc. that make comparison with XFS and others quite favorable. We know that ext3 cannot go fast enough, but it is not used with Lustre. Those changes are currently being merged upstream btw. - Peter - ________________________________ From: lustre-discuss-bounces@clusterfs.com [mailto:lustre-discuss-bounces@clusterfs.com] On Behalf Of ??? Sent: Wednesday, May 24, 2006 7:53 AM To: lustre-discuss@clusterfs.com Subject: [Lustre-discuss] degraded performance on lustre! hi, Does someone compare performance of ext3 with xfs? I have suffer extreme performance degrade when writing to ext3, about 3 times slower than xfs on same hardware.Also lustre have bad performance.My testsuite is sgpdd_survey and obdfilter_survey (CFS recommended). Our results: write read sgpdd_survey(raw device): 300MB/s 390MB/s obdfiler_survey(lustre with one ost): 110MB/s 360MB/s iozone(xfs): 320MB/s 340MB/s iozone(ext3): 118MB/s 220MB/s Our platform: Hardware: Dual EMT64 2GMEM areca 1260 with 7 sata2 disks(raid5) OS:linux-2.6.9-22 FS: lustre-1.4.6 It seems that ext3 and lustre cann''t achive much performance from high transfer rates raid adapter.But XFS do it. Or there are some tunning to be done to improve ext3 performance? I have also sent this to lustre bugzilla. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mail.clusterfs.com/pipermail/lustre-discuss/attachments/20060524/753323c6/attachment-0001.html
hi, Does someone compare performance of ext3 with xfs? I have suffer *extreme* performance degrade when writing to ext3, about 3 times slower than xfs on same hardware.Also lustre have bad performance.My testsuite is sgpdd_survey and obdfilter_survey (CFS recommended). Our results: write read sgpdd_survey(raw device): 300MB/s 390MB/s obdfiler_survey(lustre with one ost): 110MB/s 360MB/s iozone(xfs): 320MB/s 340MB/s iozone(ext3): 118MB/s 220MB/s Our platform: Hardware: Dual EMT64 2GMEM areca 1260 with 7 sata2 disks(raid5) OS:linux-2.6.9-22 FS: lustre-1.4.6 It seems that ext3 and lustre cann''t achive much performance from high transfer rates raid adapter.But XFS do it. Or there are some tunning to be done to improve ext3 performance? I have also sent this to lustre bugzilla. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mail.clusterfs.com/pipermail/lustre-discuss/attachments/20060524/822868b6/attachment-0001.html
Hi, Is there in fact a reason that mballoc and extents are not format options? - Peter -> -----Original Message----- > From: lustre-discuss-bounces@clusterfs.com > [mailto:lustre-discuss-bounces@clusterfs.com] On Behalf Of Cliff White > Sent: Friday, May 26, 2006 9:51 AM > To: ?????? > Cc: lustre-discuss@clusterfs.com > Subject: Re: [Lustre-discuss] degraded performance on lustre! > > ??? wrote: > > hi, > > Does someone compare performance of ext3 with xfs? I have suffer > > /extreme/ performance degrade when writing to ext3, about 3 times > > slower than xfs on same hardware.Also lustre have bad > performance.My > > testsuite is sgpdd_survey and obdfilter_survey (CFS recommended). > > Please be sure you use ldiskfs for Lustre testing. In > addition, you should use the mballoc and extent options when > creating or mounting. > When creating lustre objects, specify '' --mountfsoptions > extents,mballoc'' > > The best way to create a lustre partition is with ''lconf > --reformat <XML > file>" - otherwise, you can create an OST filesystem with > # mke2fs -j -J size=400 -i 16384 {dev} > > If you wish to mount an ldiskfs partition for non-lustre > tests, you can do this: > mount -t ldiskfs -o extents,mballoc /dev/sdb4 /mnt/test > > Hope this helps > cliffw > > > > > Our results: > > > write > > read > > sgpdd_survey(raw device): 300MB/s 390MB/s > > obdfiler_survey(lustre with one ost): 110MB/s 360MB/s > > iozone(xfs): > 320MB/s 340MB/s > > iozone(ext3): > 118MB/s 220MB/s > > > > Our platform: > > Hardware: Dual EMT64 2GMEM areca 1260 with 7 sata2 disks(raid5) > > OS:linux-2.6.9-22 > > FS: lustre-1.4.6 > > > > It seems that ext3 and lustre cann''t achive much > performance from high > > transfer rates raid adapter.But XFS do it. Or there are > some tunning > > to be done to improve ext3 performance? > > I have also sent this to lustre bugzilla. > > > > > > > > > > > ---------------------------------------------------------------------- > > -- > > > > _______________________________________________ > > Lustre-discuss mailing list > > Lustre-discuss@clusterfs.com > > https://mail.clusterfs.com/mailman/listinfo/lustre-discuss > > _______________________________________________ > Lustre-discuss mailing list > Lustre-discuss@clusterfs.com > https://mail.clusterfs.com/mailman/listinfo/lustre-discuss > >
??? wrote:> hi, > Does someone compare performance of ext3 with xfs? I have > suffer /extreme/ performance degrade when writing to ext3, about 3 times > slower than xfs on same hardware.Also lustre have bad performance.My > testsuite is sgpdd_survey and obdfilter_survey (CFS recommended).Please be sure you use ldiskfs for Lustre testing. In addition, you should use the mballoc and extent options when creating or mounting. When creating lustre objects, specify '' --mountfsoptions extents,mballoc'' The best way to create a lustre partition is with ''lconf --reformat <XML file>" - otherwise, you can create an OST filesystem with # mke2fs -j -J size=400 -i 16384 {dev} If you wish to mount an ldiskfs partition for non-lustre tests, you can do this: mount -t ldiskfs -o extents,mballoc /dev/sdb4 /mnt/test Hope this helps cliffw> > Our results: > write > read > sgpdd_survey(raw device): 300MB/s 390MB/s > obdfiler_survey(lustre with one ost): 110MB/s 360MB/s > iozone(xfs): 320MB/s 340MB/s > iozone(ext3): 118MB/s 220MB/s > > Our platform: > Hardware: Dual EMT64 2GMEM areca 1260 with 7 sata2 disks(raid5) > OS:linux-2.6.9-22 > FS: lustre-1.4.6 > > It seems that ext3 and lustre cann''t achive much performance from high > transfer rates raid adapter.But XFS do it. Or there are some tunning to > be done to improve ext3 performance? > I have also sent this to lustre bugzilla. > > > > > ------------------------------------------------------------------------ > > _______________________________________________ > Lustre-discuss mailing list > Lustre-discuss@clusterfs.com > https://mail.clusterfs.com/mailman/listinfo/lustre-discuss