UNSOLVED

awesome2

updated

16 years ago

A

awesome2

20 Posts

0

1567

March 5th, 2011 07:00

Performance issue on vmax

Hello,

We have performance issues with our VMAX on the linux servers. I can see from SPA that the write respose times are in between 17-22 ms for about 300 to 400 iops. I lookedd at the FA ports should i consider %busy to see if there is a bottleneck at the ports?? what else should i look to find out the reason and also

I have .ttp files from vmax which are 2 hrs interval fies for a day is there a way to make those 12 files(1day) into 1 file(1day). I have a STP navigator to open these files.??

And what is Symmerge used for compared to STP navigator?? 

  • Quincy561

    1281 Posts

    668

    0

    Posted March 5th, 2011 08:00

    There is probably too many variables to solve the problem here.  I suggest that you get customer service involved.  Could be some of the IOs are larger than 256k.  If so, you should find a way to reduce the IO size.  Also could be IO bursts, or not enough logical devices or FA paths if RDF/S.  

  • awesome2

    20 Posts

    668

    0

    Posted March 5th, 2011 08:00

    Yeah  there could be many things involved here i suspect that the io sizes are about 256kb.

    But From SPA how to determine that the FA is over utillized. should i look for %busy

    And i have ECC performance manager data and STP data from vmax i analyse using STP navigator but find very different statistics for both data can you tell me which would be more accurate(performance manager or Stp DATA)??

    And what is SYmmerge used for?? 

  • awesome2

    20 Posts

    668

    0

    Posted March 5th, 2011 08:00

    I/o sizes are about 150-175kb on an average..

    They are doing SRDF to the remote site.

  • Quincy561

    1281 Posts

    668

    0

    Posted March 5th, 2011 08:00

    What is the IO size?

  • Quincy561

    1281 Posts

    668

    0

    Posted March 5th, 2011 08:00

    And is there any replication involved?

  • Quincy561

    1281 Posts

    668

    0

    Posted March 5th, 2011 09:00

    STP and ECC can only monitor IO rates at 1 minute samples, you may have write bursts that happen for a few milliseconds, then stop.  This can lead to very high utilizations for very short periods of time, and idle other times.  SPA should be the best method to monitor the system.

    SymmMerge is used to calculate utilizations, but still is relying on data at 1 minute samples.

    I would look at your queues at the host level.  If you are seeing high queues with short samples, say 5 seconds, I would suggest adding more FAs or devices.

  • RobertDudley

    2 Intern

    •

    448 Posts

    668

    0

    Posted March 8th, 2011 06:00

    you mentioned SRDF, what mode is it running in?

  • awesome2

    20 Posts

    668

    0

    Posted March 11th, 2011 07:00

    Sorry for late respose was out

    As of now SRDF/s but this is only on some devices.

    Devices which are not in any Device group(not been replicated) also show same response time 17-20 ms.

    i will try today to merge .ttp files and Thanks all