You are correct that you could have data loss, but he could still "survive" a two-drive loss IF the they were the "proper" two drives (OA+MB or OB+MA in your example). I should have stated that I didn't recommend running in this degraded mode, but the system could survive a two drive loss with RAID10 to get the system down and the problem corrected.
We've used RAID10 (mirror then stripe) both hardware and software on our production database servers for sometime now. I can tell you that you can loose multiple drives and still run in degraded mode. Obviously you wouldn't want to run in this situation in production, but it does give you time to get the hardware you need and/or shutdown the system so that no data loss occurs. For example, we have 2 Dell PowerVault 220S disk arrays with 14 drives in each array as the data area for our production Linux database server. We had a Dell PERC3/QC four-channel RAID controller connected to the PowerVaults. We had a combination of hardware RAID1 and software RAID0 because of a limitation of the hardware RAID card. Check out http://w3.one.net/~djflux/graphics/raiddiag.png for a visual (ignore the software RAID1 (/dev/md1) as we don't use that anymore, but the top portion is what we have now with all software RAID). The 220's can have a split backplane so that two SCSI controller channels (or 2 channels of a RAID controller) can control one 14-disk PV. One morning I came in and one whole side of one PowerVault was offline (7 drives) and the machine was still cranking away with no problems (other than a quarter of it's array was gone :) ). I shutdown the server, checked all the drives, replaced the SCSI cable and away I went. No data loss. I admit I was lucky but it always pays to have good backups as well ;) As I said, I don't recommend running a production system degraded like that, but from experience I know that it will stay up IF the "correct" drives go offline. My apologies for not being precise :) Talk to you later, Andy. -----Original Message----- From: Ward William E DLDN [mailto:[EMAIL PROTECTED] Sent: Thursday, March 20, 2003 3:22 PM To: '[EMAIL PROTECTED]' Subject: RE: RAID modes and priority > -----Original Message----- > From: Rechenberg, Andrew > [mailto:[EMAIL PROTECTED] > Sent: Thursday, March 20, 2003 9:12 AM > To: [EMAIL PROTECTED] > Cc: [EMAIL PROTECTED] > Subject: RE: RAID modes and priority > > > RAID10 can theoretically survive a two drive loss in your siutation > (they have to be the correct two drives :) ), whereas RAID0+1 will die > if you have a two drive loss. Andrew, that can't be right in the situation that Peter was describing. Raid 0+1 is striping and mirroring, vice RAID 10 being mirroring and striping. In a 4 drive setup, that means he has a stripe across two drives, and then is mirroring those two drives to the other two. Like this: OA-OB STRIPE CHAIN | | MIRROR MA-MB STRIPE CHAIN With O being the "original" and M being the "mirror" sets. Obviously, you can lose any ONE drive without loss of data in this setup, since you have a totally intact stripe on either the Original side or the Mirror side. Depending on how you implement the RAID unit (and whether it's hot swappable, for example) you may not even have to go down, although that's highly dependent on your particular software RAID controller. Now, if you lose TWO drives at one time, you have two distinct possibilities. You can lose either an entire stripe (either the original or the mirror) or you can lose one drive off of each. Again, if you lose an entire stripe, obviously lose nothing, and again, depending on your setup, you may not even go down, though you will obviously fail down to a RAID 0. It gets more dangerous, however, when you lose a drive off each stripe... Let's say you lose OA and MB (or MA and OB, doesn't matter). In that case, you're going to have some minor problems, depending on your RAID controller. In this case, you should be able to safely shut down, and you wouldn't lose any data, but you DEFINITELY don't want to run in this configuration. Most Software RAID units won't start in this configuration, but some should be able to RUN in this configuration long enough to shutdown. You could be back up and operable (albeit without mirroring, so defaulting back down to RAID 0), within minutes, though, and it may be as simple as a change in the RAID lashing in a the RAID BIOS on some software RAID, such as the ones on some motherboards, without a need to go in and physically switch the cables around. The WORST case is if you lose both "pairs" of drives, the original and the mirror (OA and MA or OB and MB). You have lost data, and CANNOT recover in this case; better pull out your backups. Even worse, you won't be able to recover much (depending on the size of the stripes) even from the existing drives; if a file isn't COMPLETELY on one drive (the surviving drive) then it's unrecoverable. However, it's pretty obvious that even in 2 drive failure mode, that should only happen in 1 our of three cases. And 2 drive failure mode should be pretty uncommon to begin with, barring a SPECIFIC event, in which case you have other problems, anyway; lightening strikes, power surges, etc., should be engineered out at the beginning, via UPS and line conditioning. RAID 0+1 does not guarantee under all possible two drive failure modes that you don't lose data, but it's still fairly robust even in that mode. No RAID unit can support 1/2 of the drives being removed under all regimens, either (except RAID 1, in a 2 drive setup, but I don't think we're really talking about just mirroring, are we?). RAID10 and RAID 0+1 are (in a way) opposites, but in both cases, they should behave somewhat similarly under this particular failure mode. Of course, you could ignore everything I've said; all of the RAID units I've ever built were RAID 5 (although that's changing this weekend, as I'm going to be setting up RAID0 on both my wife's and my computers (CompUSA had 7200RPM Maxtor 120GB ATA-133 HDD for $79.95 after mail in rebate... sale ends today, for folks who might suddenly be saying "Hmmmm" ;) )so my experience with RAID 0, RAID 1, RAID10, and RAID 0+1 is all theoretical. I've built plenty of RAID 5 units, though. -- redhat-list mailing list unsubscribe mailto:[EMAIL PROTECTED] https://listman.redhat.com/mailman/listinfo/redhat-list -- redhat-list mailing list unsubscribe mailto:[EMAIL PROTECTED] https://listman.redhat.com/mailman/listinfo/redhat-list
