OES11 SP2 Performance


We have a tree with OES11SP1/SLES11SP1 nodes. We recently added nodes
with OES11SP2/SLES11SP3 nodes. The performance on these nodes was
noticeably poorer. The only changes we made to both types of nodes was
increasing the cache from the default of 2GB to 8GB. We see a 99% cache
hit ratio on both. These are also virtual servers. The older ones that
perform well are running OES11SP1 and have 10GB of RAM and 4 CPUs
running on Windows 2008R2. The newer ones that perform poorly are
running OES11SP2 and have 16GB of RAM and 8 CPUs running on Windows
2012R2. I know that there is a alot of tuning that can be done to eDir,
but we never did that on SP1. The best way to get the load needed to
test and see the poor performance is to the the servers into production
which is not always an option.

Is there some major change to OES11SP2/SLES11SP3 that would cause this
performance change? Something common that I should change?


--
sfoley
------------------------------------------------------------------------
sfoley's Profile: https://forums.netiq.com/member.php?userid=9549
View this thread: https://forums.netiq.com/showthread.php?t=53370

Tags:

Parents
  • On 04/23/2015 10:44 AM, sfoley wrote:
    >
    > We have a tree with OES11SP1/SLES11SP1 nodes. We recently added nodes
    > with OES11SP2/SLES11SP3 nodes. The performance on these nodes was
    > noticeably poorer. The only changes we made to both types of nodes was


    How is the performance being measured, exactly, and what is being done to
    cause the load being measured?

    > increasing the cache from the default of 2GB to 8GB. We see a 99% cache


    The DIB cache default with 8.8 SP8 is 200 MB, not 2 GB; this has been the
    default, statically allocated, since (as I recall) 8.8 SP5. If you had a
    different default (likely dynamically allocated) then it was from a
    version of eDirectory upgraded from an older version.

    > hit ratio on both. These are also virtual servers. The older ones that


    Having 99% hits from cache is definitely good.

    > perform well are running OES11SP1 and have 10GB of RAM and 4 CPUs
    > running on Windows 2008R2. The newer ones that perform poorly are
    > running OES11SP2 and have 16GB of RAM and 8 CPUs running on Windows
    > 2012R2. I know that there is a alot of tuning that can be done to eDir,
    > but we never did that on SP1. The best way to get the load needed to
    > test and see the poor performance is to the the servers into production
    > which is not always an option.


    8.8 SP8 introduced a lot of performance enhancements that were very
    noticeable, depending on what you are doing, but that does not mean your
    tests are testing those areas. Also, when you add a new server into the
    tree some things, which are almost entirely unrelated to performance and
    the cache, may be set differently on the new box, such as indexes defined
    for the server. A big one is the 'Object Class' index, specifically the
    value index, which I think should be a default with eDirectory always.
    Another possible difference is that eDirectory will not automatically add
    replicas of things if it is the fourth (or higher) server in a partition,
    so any queries will now need to go and chat with other servers,
    introducing delays.

    --
    Good luck.

    If you find this post helpful and are logged into the web interface,
    show your appreciation and click on the star below...
Reply
  • On 04/23/2015 10:44 AM, sfoley wrote:
    >
    > We have a tree with OES11SP1/SLES11SP1 nodes. We recently added nodes
    > with OES11SP2/SLES11SP3 nodes. The performance on these nodes was
    > noticeably poorer. The only changes we made to both types of nodes was


    How is the performance being measured, exactly, and what is being done to
    cause the load being measured?

    > increasing the cache from the default of 2GB to 8GB. We see a 99% cache


    The DIB cache default with 8.8 SP8 is 200 MB, not 2 GB; this has been the
    default, statically allocated, since (as I recall) 8.8 SP5. If you had a
    different default (likely dynamically allocated) then it was from a
    version of eDirectory upgraded from an older version.

    > hit ratio on both. These are also virtual servers. The older ones that


    Having 99% hits from cache is definitely good.

    > perform well are running OES11SP1 and have 10GB of RAM and 4 CPUs
    > running on Windows 2008R2. The newer ones that perform poorly are
    > running OES11SP2 and have 16GB of RAM and 8 CPUs running on Windows
    > 2012R2. I know that there is a alot of tuning that can be done to eDir,
    > but we never did that on SP1. The best way to get the load needed to
    > test and see the poor performance is to the the servers into production
    > which is not always an option.


    8.8 SP8 introduced a lot of performance enhancements that were very
    noticeable, depending on what you are doing, but that does not mean your
    tests are testing those areas. Also, when you add a new server into the
    tree some things, which are almost entirely unrelated to performance and
    the cache, may be set differently on the new box, such as indexes defined
    for the server. A big one is the 'Object Class' index, specifically the
    value index, which I think should be a default with eDirectory always.
    Another possible difference is that eDirectory will not automatically add
    replicas of things if it is the fourth (or higher) server in a partition,
    so any queries will now need to go and chat with other servers,
    introducing delays.

    --
    Good luck.

    If you find this post helpful and are logged into the web interface,
    show your appreciation and click on the star below...
Children
No Data