September 22, 2005 at 9:16 am
I have a cluster of SQL Servers in Active/Passive mode and when I force a Move Group or fail it over, the SQL Network Name takes a while to come back online.
I realize that the whole cluster won't move instantaneously but I'm trying to make the failover happen as quickly as possible.
Does anyone have any suggestions as to how to speed up the failover process? Are there any good practices for configuring the cluster items outside of the default settings?
September 23, 2005 at 2:45 am
Have you had a look at your WINS server?
September 23, 2005 at 3:18 am
MS Clusters are MS Clusters, and there's not much more u can say! When a failover occurs, there's usually so much it has to do that some single thing may appear to take an age. When u add up all the domain security, DNS (or WINS i suppose), state transfer, disk transfer etc, etc, i'm always quite amazed that it even manages to failover at all ... let alone quickly!? Unless ur talking minutes upon minutes to failover, i wudn't worry ... and do try to remember that failing over isn't something thats suppose to happen often!!
As for even attempting to try and go lower level to try and speed it up, i'm guessing this is ur first cluster .... otherwise the thought wud never have entered ur head!
Just stare at the box(s) like so , and marvel at the wonder of ur rundundant (albeit, slightly slow) system!!
September 23, 2005 at 10:31 am
What version and SP is your cluster OS ? We presently have a few active/passive clusters running SQL 2KSP3a. We upgraded most of them from WIN 2K/SP4 to Win 2K3/SP1 and noticed a dramatic difference in failover. However you still have to wait about 30 seconds for the network name to come online (better than the 60-90 seconds in Win 2K).
There are DNS/Wins/AD type of things that can be done to help, but the gains are not worth the effort unless you are processing huge tranasctinal volumes over a farm of SQL Servers - I know because I once did extensive SQL Server and Cluster work at an internet stock trading company.
RegardsRudy KomacsarSenior Database Administrator"Ave Caesar! - Morituri te salutamus."
September 23, 2005 at 11:16 am
Right now I am running W2K3 with SQL 2KSP4.
Actually I am not running WINS at all. The cluster is running in a small domain with dns. Does adding a WINS server help with the network name? I was under the assumption that DNS took care of that.
September 23, 2005 at 11:26 am
I do not know if adding WINs would make a difference. We use DNS and WINs as standards here.
How much time are you talking abut having to wait anyway ?
RegardsRudy KomacsarSenior Database Administrator"Ave Caesar! - Morituri te salutamus."
September 23, 2005 at 11:31 am
Full failover takes ~5 minutes.
September 23, 2005 at 12:37 pm
Failover has various phases: IP address, storage, services then SQL Server startup. In a Win 2K3/SP1 w/SQL 2K/SP3, I can see a complete 'failover' taking 3-5 minutes. Our clusters generally are in the 2-3 minute range. You may be getting stuck on SQL Server if you have a large number of databases on the cluster because SQL Server does not let you in until every database has been recovered. In SQL 2K5 the SQL startup will allow access to each database as they are recovered instead of waiting for all of the databases to be recovered. So, the bottom line is that you may be just a bit on the 'hig'h side server response to a 'failover'. You have to read the fine print ...
Oh, don't think that switching hardware and operating system will help either. I've worked on failover solutions on Sun Solaris and HP with Sybase and Oracle as the DBMSs. The best time achieved was on Sun Solaris with Sybase - 3 minutes !
RegardsRudy KomacsarSenior Database Administrator"Ave Caesar! - Morituri te salutamus."
Viewing 8 posts - 1 through 7 (of 7 total)
You must be logged in to reply to this topic. Login to reply