Post reply

Adding Filegroups

  • June 25, 2009 at 5:47 am

    #1015226

    Paul White (6/24/2009)

    Carl Federl (6/24/2009)

    I think if you look further, you will find Cores 3 and 7 are handling all of the network IO , hence the higher CPU usage. There is a good write up titled "Scaling Heavy Network Traffic with Windows" at http://blogs.msdn.com/sqlcat/archive/2008/09/18/scaling-heavy-network-traffic-with-windows.aspx

    The recommendation is to bind the NICs to particular cores using the Interrupt-Affinity Policy Tool and then set SQL Server to not use those CPUs.

    An important caveat is that this advice is really aimed at 32-bit Windows 2000 and Windows 2003 RTM.

    These were unable to spread network-related CPU load over multiple cores.

    It is true that setting network adapter affinity may benefit very high-end systems with very high network I/O (think 1Gb/s and up) and very many cores - say 12 to 16 at the least. I doubt it applies here - though perhaps Apollo would confirm the network traffic level.

    The higher (though still comfortable) CPU utilization on those two cores could be explained by a couple of long-running MAXDOP 1 tasks, or perhaps I/O is affinitized. Who knows.

    Paul

    The bytes total/sec on 1 NIC was averaging about 20 MB.

  • June 29, 2009 at 4:53 am

    #1016712

    Just wanted to give you all an update since this forum was very instrumental in helping us choose the right direction.

    We chose to not add any other LUNS and we couldn't separate the LOG file from the Database file because this client has chosen to use snap restore as it's method of backup and recovery. According to the NETAPP documentation, in order to use snap restore, each file has to reside on the same LUN. Why is beyond me but I'll eventually figure it out.

    The good thing is after all of the work we put into optimizing queries and indexes, it seems that our counter values have significantly improved. The average page life expectancy for last week was at 899. The disk time on the primary LUN was down to 127 and the average lock time (ms) was down to an average of 30.

    We are going to look into adding a third file to tempdb. The tempdb files are on separate disks (raid 10) and we're thinking about setting MAXDOP back to 0 to see how the system behaves.

    Thanks

    James

Viewing 2 posts - 16 through 16 (of 16 total)

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