Hi, we are in the process of configuring a thumper (Sun X4500) running RHEL 4.5 as an OSS/OST - for this we need a patch for the Sun-Marvell driver to disable the disk write cache - did anybody has this patch ??? best regards, Martin
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA512 Hi Martin, Are you using the stock sata_mv driver that uses libata or the proprietary marvell driver v340? We have been using both of them, both with their fair share of problems. I have seen extremely poor performance from the sata_mv driver, and the sdparm commands for turning off WCE are not honored. With the proprietary mv_sata driver the performance is pretty good under RHEL4, but also does not seem to honor sdparm commands for disabling WCE bit. After doing a little work to port this driver to RHEL5 the driver looks to be completely broken. Currently we do not have a patch for disabling the WCE bit for either driver. IMHO I think it would be best to try and improve the sata_mv driver in the RHEL kernel since it seems the proprietary driver has been abandoned. If anyone else knows of another linux project for the Marvell driver I would be happy to hear about it. - -Herb Martin Gasthuber wrote:> Hi, > > we are in the process of configuring a thumper (Sun X4500) running > RHEL 4.5 as an OSS/OST - for this we need a patch for the Sun-Marvell > driver to disable the disk write cache - did anybody has this patch ??? > > best regards, > Martin > > _______________________________________________ > Lustre-discuss mailing list > Lustre-discuss at clusterfs.com > https://mail.clusterfs.com/mailman/listinfo/lustre-discuss-----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.7 (GNU/Linux) Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org iD8DBQFHKhmOP/62XqEEbMYRChJBAJ9cJltz29FqXxJYJ1uaSRsULG5MdgCeITqf XZR7yjCl4tk2ZkqpjUxmgko=IjDc -----END PGP SIGNATURE-----
Hi I have just filed the patch for disabling the write cache for mv_sata. Please see below. https://bugzilla.clusterfs.com/show_bug.cgi?id=14040 Thanks, -Ihara Martin Gasthuber wrote:> Hi, > > we are in the process of configuring a thumper (Sun X4500) running > RHEL 4.5 as an OSS/OST - for this we need a patch for the Sun-Marvell > driver to disable the disk write cache - did anybody has this patch ??? > > best regards, > Martin > > _______________________________________________ > Lustre-discuss mailing list > Lustre-discuss at clusterfs.com > https://mail.clusterfs.com/mailman/listinfo/lustre-discuss
On Nov 02, 2007 10:32 +0900, Syuuichi Ihara wrote:> I have just filed the patch for disabling the write cache for mv_sata. > Please see below. > https://bugzilla.clusterfs.com/show_bug.cgi?id=14040Ihara-san, have you looked at submitting this patch to the upstream driver maintainers and/or Red Hat? It looks like the patch is hard coded to disable the write cache, but it would be desirable to allow this to be a tunable parameter (e.g. module parameter, or /proc setting?)> Martin Gasthuber wrote: > > we are in the process of configuring a thumper (Sun X4500) running > > RHEL 4.5 as an OSS/OST - for this we need a patch for the Sun-Marvell > > driver to disable the disk write cache - did anybody has this patch ???Cheers, Andreas -- Andreas Dilger Sr. Software Engineer, Lustre Group Sun Microsystems of Canada, Inc.
Hi Andreas, Yes, I absolutely agree. I will file this as RFE. But, the patch is useful now as work-around for disabling the write cache of all of drives. Thanks, -Ihara Andreas Dilger wrote:> On Nov 02, 2007 10:32 +0900, Syuuichi Ihara wrote: >> I have just filed the patch for disabling the write cache for mv_sata. >> Please see below. >> https://bugzilla.clusterfs.com/show_bug.cgi?id=14040 > > Ihara-san, > have you looked at submitting this patch to the upstream driver maintainers > and/or Red Hat? It looks like the patch is hard coded to disable the write > cache, but it would be desirable to allow this to be a tunable parameter > (e.g. module parameter, or /proc setting?) > >> Martin Gasthuber wrote: >>> we are in the process of configuring a thumper (Sun X4500) running >>> RHEL 4.5 as an OSS/OST - for this we need a patch for the Sun-Marvell >>> driver to disable the disk write cache - did anybody has this patch ??? > > Cheers, Andreas > -- > Andreas Dilger > Sr. Software Engineer, Lustre Group > Sun Microsystems of Canada, Inc. > > _______________________________________________ > Lustre-discuss mailing list > Lustre-discuss at clusterfs.com > https://mail.clusterfs.com/mailman/listinfo/lustre-discuss
On Fri, Nov 02, 2007 at 10:11:10AM +0800, Andreas Dilger wrote:> have you looked at submitting this patch to the upstream driver maintainers > and/or Red Hat? It looks like the patch is hard coded to disable the write > cache, but it would be desirable to allow this to be a tunable parameter > (e.g. module parameter, or /proc setting?)Actually, this patch is against the out-of-tree Marvell driver (mv_sata), not the in-kernel driver (sata_mv.) Why are you still using this driver? You should really be using the in-kernel driver (and fixing it if needed) since mv_sata is abit of a dead end. Cheers, Jody> > Martin Gasthuber wrote: > > > we are in the process of configuring a thumper (Sun X4500) running > > > RHEL 4.5 as an OSS/OST - for this we need a patch for the Sun-Marvell > > > driver to disable the disk write cache - did anybody has this patch ??? > > Cheers, Andreas > -- > Andreas Dilger > Sr. Software Engineer, Lustre Group > Sun Microsystems of Canada, Inc. > > _______________________________________________ > Lustre-discuss mailing list > Lustre-discuss at clusterfs.com > https://mail.clusterfs.com/mailman/listinfo/lustre-discuss--