iSCSI performance issues - Storage

This is a discussion on iSCSI performance issues - Storage ; Hello technicians with clue, we have some weird and unworkable issues since we migrated a fileserver to new hardware. The choice been made to centralize the data by using an iscsi solution and existing network hardware. The hardware exists of ...

+ Reply to Thread
Results 1 to 2 of 2

Thread: iSCSI performance issues

  1. iSCSI performance issues

    Hello technicians with clue,

    we have some weird and unworkable issues since we migrated a
    fileserver to new hardware.
    The choice been made to centralize the data by using an iscsi solution
    and existing network hardware. The hardware exists of a Dell NX1950
    directly connected to a Dell MD3000 storage cabinet. The NX1940 has 2
    quad cpu's and 4 GB memory, 4 NICs each pair is teamed, one for the
    data network and one for the iscsi vlan. This NX1950 runs MS UDS
    server and one target is created which provides two virtual disks.

    The fileserver is the initiator and is a brandnew PE2950 with 2 dual
    core cpu's and 2 NICs. One for the data network and one for the iscsi
    vlan. The iSCSI initiator is the latest version 2.05 and the two disks
    are mounted successfully. I did not test the performance by iometer or
    any other tool, but just by copying large amounts of files from
    various sources. This looked promising, the speed was no different as
    with physically attached drives.

    See: http://img27.picoodle.com/img/img27/...gm_cea6420.gif
    for a picture.

    Now around 100 users connecting to their shares and all having their
    outlook PST files on their home shares (these PST's are around 2GB). I
    know it is not supported and using PST's this way only causes overhead
    and problems, but this was not the case on the old fileserver.
    The problems we are seeing is that users lose the connection with the
    PST file. And that users using shared MS Access databases are
    experiencing the thing. And that some users get the message that there
    is not enough space left on a share while writing some big files,
    while there is plenty left.

    Also a weird behaviour is that connecting with RDP to this fileserver
    is like a lottery, after trying 4,5 or 6 times connection can be made,
    else RDP client just dissapears.

    Now I have checked with perfmon and taskmanager, and other tools just
    to find the cause, but nothing helped. tcp filtering (wireshark)
    showed there are some "NBSS data continuation data" packets, but I
    suggest this is normal behaviour.

    Winperf showed that the "avg disk queue lenght" is spiking for the
    mounted disks, pages/sec is spiking in relation with the above
    counter. Using the iSCSI session counters for Bytes "received/send"
    shows continueal 100% load.

    What is also odd that I noticed, is that when I do a copy of some
    large files from or to the iSCSI connected shares the iSCSI network
    interface increases to 20, 30 and even 50%. in normal operation,
    server used by 100 users (with all the issues we have, outlook pst
    etc) this interface doesn't go further than 4 or 5%. Very strange.
    Looks like the bottleneck is somewhere else or somthing is configured
    totally wrong.

    I'm no expert i/o performance monitoring, really don't know nothing
    about it. And after 2 days of searching I give up and beg for help
    here. If someone could point me in a right direction, then please.


    Also as a side issue, after I rebooted the fileserver I lost the
    sharing on the folders. Is this normal behaviour after server loses
    it's disks?


    Thanks in advance, relikwie


  2. Re: iSCSI performance issues

    First thing I would do is lose the NIC Teaming...even in iSCSI clusters it
    is not supported -
    http://www.microsoft.com/windowsserv...sicluster.mspx

    Also ensure the NICs are GigE minimum...which they probably are, but I felt
    a need to mention that.

    Use MPIO instead for fault tolerance.

    Test without the Teaming in Place and see what your results are.


    --
    Chuck Timon, Jr.
    Microsoft Corporation
    Windows Server 2008 Readiness Team
    This posting is provided 'AS IS" with no warranties, and confers no rights.
    "relikwie" wrote in message
    news:1195076370.428809.196880@19g2000hsx.googlegro ups.com...
    > Hello technicians with clue,
    >
    > we have some weird and unworkable issues since we migrated a
    > fileserver to new hardware.
    > The choice been made to centralize the data by using an iscsi solution
    > and existing network hardware. The hardware exists of a Dell NX1950
    > directly connected to a Dell MD3000 storage cabinet. The NX1940 has 2
    > quad cpu's and 4 GB memory, 4 NICs each pair is teamed, one for the
    > data network and one for the iscsi vlan. This NX1950 runs MS UDS
    > server and one target is created which provides two virtual disks.
    >
    > The fileserver is the initiator and is a brandnew PE2950 with 2 dual
    > core cpu's and 2 NICs. One for the data network and one for the iscsi
    > vlan. The iSCSI initiator is the latest version 2.05 and the two disks
    > are mounted successfully. I did not test the performance by iometer or
    > any other tool, but just by copying large amounts of files from
    > various sources. This looked promising, the speed was no different as
    > with physically attached drives.
    >
    > See: http://img27.picoodle.com/img/img27/...gm_cea6420.gif
    > for a picture.
    >
    > Now around 100 users connecting to their shares and all having their
    > outlook PST files on their home shares (these PST's are around 2GB). I
    > know it is not supported and using PST's this way only causes overhead
    > and problems, but this was not the case on the old fileserver.
    > The problems we are seeing is that users lose the connection with the
    > PST file. And that users using shared MS Access databases are
    > experiencing the thing. And that some users get the message that there
    > is not enough space left on a share while writing some big files,
    > while there is plenty left.
    >
    > Also a weird behaviour is that connecting with RDP to this fileserver
    > is like a lottery, after trying 4,5 or 6 times connection can be made,
    > else RDP client just dissapears.
    >
    > Now I have checked with perfmon and taskmanager, and other tools just
    > to find the cause, but nothing helped. tcp filtering (wireshark)
    > showed there are some "NBSS data continuation data" packets, but I
    > suggest this is normal behaviour.
    >
    > Winperf showed that the "avg disk queue lenght" is spiking for the
    > mounted disks, pages/sec is spiking in relation with the above
    > counter. Using the iSCSI session counters for Bytes "received/send"
    > shows continueal 100% load.
    >
    > What is also odd that I noticed, is that when I do a copy of some
    > large files from or to the iSCSI connected shares the iSCSI network
    > interface increases to 20, 30 and even 50%. in normal operation,
    > server used by 100 users (with all the issues we have, outlook pst
    > etc) this interface doesn't go further than 4 or 5%. Very strange.
    > Looks like the bottleneck is somewhere else or somthing is configured
    > totally wrong.
    >
    > I'm no expert i/o performance monitoring, really don't know nothing
    > about it. And after 2 days of searching I give up and beg for help
    > here. If someone could point me in a right direction, then please.
    >
    >
    > Also as a side issue, after I rebooted the fileserver I lost the
    > sharing on the folders. Is this normal behaviour after server loses
    > it's disks?
    >
    >
    > Thanks in advance, relikwie
    >



+ Reply to Thread