Our 48 CPU SQL Server Hardware goes to 100% Utilization

  • We have an application with hundreds of users running on database hardware that has 48 Logical CPUs (HP Itanium). The server has 4 quad core processors that are hyper threaded. We rarely go over 30-40% overall utilization. Yesterday, all 48 CPUs pegged for approximately 30 minutes and then they went back down to normal. Has anyone ever seen any behavior like this before? Any suggestions as to how to "investigate" this type of incident?

  • Well there is alot you left out.

    1. What operating system you running

    2. raid configuration

    3. any other software on the system besides SQL

    4. last time system as updated

    5. did you check the task manager at this time to see what was sppiking the cPU's something

    would have been listed at the top

  • Here are the facts:

    1. What operating system you running - we ar running Windows 2003 Enterprise

    2. raid configuration - We are on a Clarion cx400 with 48 spindles configured Raid 5.

    3. any other software on the system besides SQL - No

    4. last time system as updated - OS Updaes were nearly 3 months ago. SQL Server 2005 updates when SP3 came out.

    5. did you check the task manager at this time to see what was sppiking the cPU's something

    would have been listed at the top - The top of the list was Sqlserv.exe.

    We also checked the application and there were no updates formoe than a week and the last update applied was formatting of some UI screens.

  • well if sqlserver.exe was at the top, the first thing i can think of is re-indexing.

    then you have to look at your partitions and a defrag running.

  • Check the default trace for that time. See what might have been running.

    SQL will use all resources if they are needed, so I'd start by trying to determine what might be running. Some of the backup software goes parallel as well, though I'd be surprised to have it go 48 threads.

    Also, hyperthreading used to cause worse performance in some situations. Not sure if that applies to Itanium, but I'd double check that.

  • Thanks Steve -

    There were over 1500 user sessions running at that time txtPost_CommentEmoticon(':w00t:'); so it might take a while. I have already confirmed there were no backups running. I thought of the default trace and I am crawling thru that now.

    ... Jerry

Viewing 6 posts - 1 through 5 (of 5 total)

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