--=-12UJUa0bX8JAbF7dEqsL Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: quoted-printable A Qua, 2004-04-21 =C3=A0s 05:23, Phil Schwan escreveu:> On Tue, 2004-04-20 at 17:52, Bernard Dugas wrote: > > Does Lustre manage a data redundancy insuring data conservation even in=20 > > the case of a disk crash, ie OST crash in Lustre ? ie something like=20 > > raid1 or RAID5... >=20 > Lustre stores its data on any block device, so you can run software or > hardware RAID1 or RAID5 underneath Lustre, to protect against individual > drive failures. >=20One other alternative is drbd (http://www.drbd.org). But setting it up with lustre has some interesting issues. --=20 Jo=C3=A3o Miguel Neves --=-12UJUa0bX8JAbF7dEqsL Content-Type: application/pgp-signature; name=signature.asc Content-Description: Esta =?ISO-8859-1?Q?=E9?= uma parte de mensagem assinada digitalmente -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.4 (GNU/Linux) iD8DBQBAhjEXGFkMfesLN9wRAuH4AJ4ufwfWvty4zIDhFc9kJU+w4wNNagCfWlj5 LCF20Xk1/yqawVjiyrcrR5k=Xt3w -----END PGP SIGNATURE----- --=-12UJUa0bX8JAbF7dEqsL--
Hi, João Miguel Neves wrote:> One other alternative is drbd (http://www.drbd.org). But setting it up > with lustre has some interesting issues.Which issues ? Best regards, -- Bernard Dugas +33 615 333 770
Hi Phil, Phil Schwan wrote:> We''ll incorporate a RAID1 OST feature in a future version of Lustre. It > will do duplicate writes to multiple OSTs, run in degraded mode when an > OST is down, and recover by syncing only those files which have changed.This is exactly the needs i''ve come up after some thinking in this field. Raid5 would not be bad neither. If you know anything close to that, let me know :-) Best regards, -- Bernard Dugas +33 615 333 770
--=-lNhZjaXNDBLIvrDBHjTB Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: quoted-printable A Dom, 2004-04-25 =C3=A0s 22:54, Bernard Dugas escreveu:> Jo=C3=A3o Miguel Neves wrote: > > One other alternative is drbd (http://www.drbd.org). But setting it up > > with lustre has some interesting issues. >=20 > Which issues ?They are mostly of the sort of "I''m not really sure this will always work" and "This mostly works": 1) I''m using a simple script for upcall in lustre, so that it recovers from timeouts automatically. Unfortunately that makes it more difficult to be sure that a node is disconnected from the cluster, particularly if it is a temporary network issue (you can''t automatically reach the host to be sure that lustre is disabled on the host).=20 I''m still trying to find the best way to synchronise changes from Secondary to Primary in drbd with lustre''s node replacing. And changes from Primary to Secondary with disabling lustre on the OST before it tries to connect to the MDS again. 2) DRBD has this annoying habit of syncing all the disk after a crash. I''ve already experienced some corruption issues because the host it was syncing from crashed in the middle. But that''s a drbd only issue. Some notes: - See Robert Read''s e-mail on replacing a node: https://lists.clusterfs.com/pipermail/lustre-discuss/2004-March/000200.html - https://wiki.clusterfs.com/lustre/ClientUpcall seems something I must read more thoroughly. --=20 Jo=C3=A3o Miguel Neves --=-lNhZjaXNDBLIvrDBHjTB Content-Type: application/pgp-signature; name=signature.asc Content-Description: Esta =?ISO-8859-1?Q?=E9?= uma parte de mensagem assinada digitalmente -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.4 (GNU/Linux) iD8DBQBAnJ/RGFkMfesLN9wRArB6AKCHHfNCa6sqFtQ+zi/yoIBRm3Yt7wCghFi5 wVQ6cZlJ9PuuC29Ig75peuU=j0MG -----END PGP SIGNATURE----- --=-lNhZjaXNDBLIvrDBHjTB--
Hi Bernard-- On Tue, 2004-04-20 at 17:52, Bernard Dugas wrote:> > I''ve tried to read your presentations, but have not found an answer to > my main question : > > Does Lustre manage a data redundancy insuring data conservation even in > the case of a disk crash, ie OST crash in Lustre ? ie something like > raid1 or RAID5...Lustre stores its data on any block device, so you can run software or hardware RAID1 or RAID5 underneath Lustre, to protect against individual drive failures. We''ll incorporate a RAID1 OST feature in a future version of Lustre. It will do duplicate writes to multiple OSTs, run in degraded mode when an OST is down, and recover by syncing only those files which have changed. -Phil