Please start any new threads on our new
site at https://forums.sqlteam.com. We've got lots of great SQL Server
experts to answer whatever question you can come up with.
| Author |
Topic |
|
bogey
Posting Yak Master
166 Posts |
 Posted - 2006-11-07 : 10:19:05
|
| We are currently running SQL 2005 on 3 seperate computers. Each computer has multiple databases that are critical. I've submitted a request that each computer should have it's own failover and we should not put all 3 failover's on one computer. For me administration is much easier and just in case the unthinkable happens (all 3 prod machines go down) we are then sitting on one failover until we get the prods back up.Any one like to comment on this, please do.thanks. |
|
|
Kristen
Test
22859 Posts |
Posted - 2006-11-07 : 10:30:24
|
I presume that the failover machine is carrying a copy of the databases - restored from log backups every 10 - 15 minutes, or overnight, or somesuch.Thus the machine has to have enough capacity to "hold" all the databases from all 3 production machines. I don't have any SQL boxes that are only spec'd to 1/3rd of their capacity!, so I presume therefore that its going to be a "big machine" - relative to the 3 machines its going to be capable of failing-over from.So then it is going to have a different hardware spec to the production machines, and there in lies a source of weakness.The alternative is that it is an exact replica of the hardware for EACH of the 3 production machines, and the plan is to ONLY be able to support a single failure at any one time, and then:You have NO current database restored onto the failover machine - because you don't know which machine is going to fail  After a disaster the time to restore the backups onto the failover machine is going to be unacceptableThe process of testing the failover of each of the 3 production servers (which you need to do regularly, right?) is going to be expensive and painful (for the users that will prove and sign off the failover test)So I think I'm talking myself into either:1) 3 failover machinesor2) 1 machine 3-times the size of a production machine, all databases [from all 3 production servers] regularly restored onto the machine, and this leads to a QA process completely independent from that used on Production for every patch, upgrade and release (that's SQL Server patches, Windows Updates, and any application change).I'm not liking option (2) very much!Kristen |
 |
|
|
Sitka
Aged Yak Warrior
571 Posts |
Posted - 2006-11-07 : 12:28:04
|
| prod server 1 goes downremove prod server 1 from domain (DNS entry and computer account)remove back up server from domain (via that server itself join workgroup)change IP address of back up server to prod server 1rename back up server to <prod server 1>join former backup server to domainshut down log shipping from prod server 2 and 3 for a short period of time (but still do backups)buy new prod server 1, bring online (using reverse of domain membership operations)resync log shiping back to back up server.It's doable, and a pretty good approach within the confines of a single back up machine.You would be messed with two or three failures unless you can reconfigure applicationsto point to different servers then there is some flexibility, and that limit may combine witha loss of failover response as well.Depends a lot on what causes the failure, a lone gunman may take out 3 servers just as easy as 1,if they are in the same rack. If all servers are in different buildings you would be better off.Lots of scenerios. Maybe a 3/2 solution would be best, 3 live going to 1 failover (warm swap), then an installed, unpluggedserver waiting to accept full restores (cold swap). Keeps you in the four total licence count. 3 proc + 1 cheap cal based"it's definitely useless and maybe harmful". |
|
|
|
Kristen
Test
22859 Posts |
Posted - 2006-11-07 : 12:34:55
|
| I didn't think you needed an extra license for a fail-over server, do you?Kristen |
|
|
|
Sitka
Aged Yak Warrior
571 Posts |
Posted - 2006-11-07 : 12:59:33
|
| Not sure, I always read it as you did, but I could be wrong,"it's definitely useless and maybe harmful". |
|
|
|
|
|
|
|
|
Put All failover datatabases on "1" machine