How do you cluster 2 servers for SQL

  • We just got 2 new Dell Rack Servers with 5 drive SCSI on board raid. We also have a PowerVault raid that we are going to connect to them. How do you cluster them? Both the servers have those thin long connectors for the heartbeat cable i guess.

  • Brian Knight wrote an excellent multi-part clustering article. Check it out, starting with http://www.sqlservercentral.com/columnists/bknight/clustering_a_sql_server_machine_.asp and then move on to the first of several step-by-step articles http://www.sqlservercentral.com/columnists/bknight/stepbystepclustering.asp

    Cheers,

    - Mark


    Cheers,
    - Mark

  • One, If you bought them recently from Dell they are usually pretty good about getting you started. Two, IF this is using all SCSI make sure you have the correct controllers, not all SCSI controllers are cluster aware.

    Wes

  • thats very interesting thanks, most of that is with 2000server, our servers have 2003 server on them.

  • real, quick on the second link, i noticed problems loading the pages.

  • I have several Dell clusters. There is an official "Dell Cluster Kit", which is a manual and a crossover ethernet cable for the heartbeat. It costs $195 so is very bad value or money, but they will support you as a cluster with it.

    You need a PERC 3/DC or PERC3/QC card in each machine for the shared storage. If you're on board RAID is built using PERC2 or PERC3/DI card (these two are equivalent) then it won't cluster on those buses. The long narrow connectors are VHDCI (very high density centronics interface) connectors and are for the SCSI connections to the PowerVault. You're PowerVault must have two contollers and must be bridged to operate in "cluster mode". This is done with a jumper on the controllers for a PV210S and a switch at the backf for a PV220S. Finally, you need at least two ethernet cards in each machine (Dell actually reccommend three). One for the public network and one for the heartbeat. You can connect this directly (hence the crossover) or though a switch. It's your choice, however, you you use a switch you must have a separate one for each network.

    For disks you need at least one raid set (at least a pair of mirrored disks) for each discrete unit on the cluster. The reccommended configuration for a SQL machine is: two as RAID 1 for Quorum; two as RAID 1 for data disk; two as RAID 1 for log disk. Some people add a RAID 5 for an on-line backup disk or put the data onto RAID 5 for cost saving.

  • thanks for all the insight into the dell clustering, very helpful, while reading some documentation, it says to not power up the servers or powervault until its all wired up. do you know why this is the case? see we turned one of the servers on by accident thinking it was another server. the raid was turned on briefly for 20 seconds. is this going to be a problem when the dell people come out to do the clustering?

  • nevermind about the powering up of the server problem. the dell guy just got here and said it isn't a problem.

  • Whats funny is I use to be "the Dell guy" worked in enterprise server space for a couple of years.

    Wes

  • the fail over wait period is something we are worried about. one might think that you can lose one server and that instant you get the passive one to take its place, but not so. it has to go the quorum and read the log and all that stuff, which takes a few minutes. alot of the programs will error out, ah well

  • All connections will be reset when the cluster to fail over. Depending on the size of the sql server how much ram how big the tempdb is will effect the speed of the failover as well. We have some clusters that failover in seconds and some that can take up to two minutes.

    Wes

  • as a test we had a directory open and then turned of the active machine, it took about a minute until the server would stop saying 'file location does not exist'

  • That's not a fair test of failing over the SQL Server, since you're using windows explorer and the lanman server process, which have lookup latencies. You should use a simple poller such as

    OSQL -Sservercluster -E -Q"exec sp_helpdb dbname"

    to check how fast your SQL server comes back.

  • ah ok, well your right. i will try this poller you gave when we get sql installed. on one of our server we got errors on it and dell is sending us new parts. it all happened when we did the 32 updates from #3 service pack

Viewing 14 posts - 1 through 13 (of 13 total)

You must be logged in to reply to this topic. Login to reply