I''m considering setting up a poor man''s cluster. The hardware I''d like to use for some critical services is especially attractive for price/space/performance reasons, however it only has a single power supply. I''m using S10 U8 and can''t migrate to OpenSolaris. It''s fine if a server dies (ie, power supply failure) and I have to manually get the service running on the other server. But what''s not ok is to lose data from the last transaction on the "primary" server. For this reason I''m reluctant to do a zfs send | zfs recv setup (besides the bad vibes I get from such a plebian method anyway). I was thinking that rather than what I''d normally do which is mirror two data drives locally, I could have one drive local and use one drive from the partner server shared via iSCSI. Comments? If that''s feasible and sane, then on the one hand it seems attractive to use a zfs backing store, but I''d have a zfs filesystem on top of the zfs backing store, which doesn''t seem helpful and I''m guessing would reduce the total space available, so I''d then have to not use the whole local disk. So I think I would want to use a raw backing store, yes? Performance is not a concern. If no one is throwing red flags yet, then to extend the idea further maybe I would move the redundant server offsite (I will have metro ethernet connectivity) and do iSCSI over the WAN and get DR for free. -frank
On Dec 15, 2009, at 11:04 PM, Frank Cusack wrote:> I''m considering setting up a poor man''s cluster. The hardware I''d > like to use for some critical services is especially attractive for > price/space/performance reasons, however it only has a single power > supply. I''m using S10 U8 and can''t migrate to OpenSolaris. > > It''s fine if a server dies (ie, power supply failure) and I have > to manually get the service running on the other server. But what''s > not ok is to lose data from the last transaction on the "primary" > server. For this reason I''m reluctant to do a zfs send | zfs recv > setup (besides the bad vibes I get from such a plebian method anyway).AVS can also be used. This one option on many of the ZFS appliances.> I was thinking that rather than what I''d normally do which is mirror > two data drives locally, I could have one drive local and use one > drive from the partner server shared via iSCSI. Comments?Seems reasonable.> If that''s feasible and sane, then on the one hand it seems attractive > to use a zfs backing store, but I''d have a zfs filesystem on top of > the zfs backing store, which doesn''t seem helpful and I''m guessing > would reduce the total space available, so I''d then have to not use > the whole local disk. So I think I would want to use a raw backing > store, yes?I would, KISS.> Performance is not a concern. > > If no one is throwing red flags yet, then to extend the idea further > maybe I would move the redundant server offsite (I will have metro > ethernet connectivity) and do iSCSI over the WAN and get DR for free.This is not DR. It is a metro cluster. DR solutions include protection against temporal failures. -- richard
On December 16, 2009 9:37:08 AM -0800 Richard Elling <richard.elling at gmail.com> wrote:> On Dec 15, 2009, at 11:04 PM, Frank Cusack wrote: > AVS can also be used. This one option on many of the ZFS appliances.I would like to but doesn''t that require OpenSolaris? I''m scared to license the StorageTek version since the Sun web page doesn''t include any pricing info.>> If no one is throwing red flags yet, then to extend the idea further >> maybe I would move the redundant server offsite (I will have metro >> ethernet connectivity) and do iSCSI over the WAN and get DR for free. > > This is not DR. It is a metro cluster. DR solutions include protection > against temporal failures.Isn''t that what backups are for? I agree using metro ethernet is not sufficient for a complete DR solution, but it does meet the offsite requirement. Or are you referring to something else? Thanks for the feedback Richard. My main concern was that zfs and iscsi didn''t seem to work all that well back in ''07 when I first looked at it. It doesn''t sound like you or anyone is concerned about that today. -frank
On Dec 16, 2009, at 10:24 AM, Frank Cusack wrote:> On December 16, 2009 9:37:08 AM -0800 Richard Elling <richard.elling at gmail.com > > wrote: >> On Dec 15, 2009, at 11:04 PM, Frank Cusack wrote: >> AVS can also be used. This one option on many of the ZFS appliances. > > I would like to but doesn''t that require OpenSolaris? I''m scared to > license the StorageTek version since the Sun web page doesn''t include > any pricing info.I read the lack of a buy-it-now button as an opportunity to negotiate :-)>>> If no one is throwing red flags yet, then to extend the idea further >>> maybe I would move the redundant server offsite (I will have metro >>> ethernet connectivity) and do iSCSI over the WAN and get DR for >>> free. >> >> This is not DR. It is a metro cluster. DR solutions include >> protection >> against temporal failures. > > Isn''t that what backups are for? I agree using metro ethernet is not > sufficient for a complete DR solution, but it does meet the offsite > requirement. Or are you referring to something else?DR is a process, not a set of products. You are describing a set of products.> Thanks for the feedback Richard. My main concern was that zfs and > iscsi > didn''t seem to work all that well back in ''07 when I first looked at > it. > It doesn''t sound like you or anyone is concerned about that today.Since you are stuck on Solaris 10, you''ll be running the old iSCSI target, with its known limitations and effective EOF. -- richard
OK, it''s been used twice now. Please closely define ''temporal failure.'' On Dec 16, 2009 1:25 PM, "Frank Cusack" <fcusack at fcusack.com> wrote: On December 16, 2009 9:37:08 AM -0800 Richard Elling < richard.elling at gmail.com> wrote:> > > On Dec 15, 2009, at 11:04 PM, Frank Cusack wrote: > > > AVS can also be used. This one option on many of the ZFS appliances. >I would like to but doesn''t that require OpenSolaris? I''m scared to license the StorageTek version since the Sun web page doesn''t include any pricing info.>> If no one is throwing red flags yet, then to extend the idea further >>maybe I would move the ... Isn''t that what backups are for? I agree using metro ethernet is not sufficient for a complete DR solution, but it does meet the offsite requirement. Or are you referring to something else? Thanks for the feedback Richard. My main concern was that zfs and iscsi didn''t seem to work all that well back in ''07 when I first looked at it. It doesn''t sound like you or anyone is concerned about that today. -frank _______________________________________________ zfs-discuss mailing list zfs-discuss at opensolaris.or... -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20091216/813526c6/attachment.html>
On Dec 16, 2009, at 6:26 PM, Jacob Ritorto wrote:> OK, it''s been used twice now. Please closely define ''temporal > failure.'' >Failure in the time domain. For example, a DBA drops a database accidentally. If the primary site and DR site are synced in time, then the DR site will also have dropped the database. OTOH, if you delay changes to the DR site by some amount of time, you can prevent the accidental drop on the DR site. There are other such examples, but in general they are of the form where something that you wanted to happen at some time didn''t, or something you didn''t want to happen did. With snapshots and ZFS, these sorts of failures can be more easily managed. -- richard