UNSOLVED

sauravrohilla

updated

11 years ago

0

1464

September 25th, 2015 12:00

VMAX Performance issues

Hello Experts,

We are trying to understand some of the performance benchmarch results that we have received from SQLIO test on windows 2012.

test scenerio: A windows VM residing on FC pool and it has three drives (C,E,F) all in the same datastore.        

C & E drives have same Allocation unit size (4K), C is MBR and E is GPT but as per SQL IO test, the throughput differs. C is slighly slower than E? Can we explain?

F drive has 64K allocation unit and gives best results which is expected.                                                                                                                                                                                                                                                                                                                                                                                            

Operating System Windows 2012 R2
Storage profile Silver
File size for SQLIO 9 GB
Disk layout:
Drive letter StartingOffset Allocation unit size Partition type
C: 1024 KB 4096 bytes (default) MBR
E: 33 MB 4096 bytes (default) GPT
F: 33 MB 64 kilobytes GPT
Benchmark results:
C: IOPS Throughput (MB/sec) E: IOPS Throughput (MB/sec) F: IOPS Throughput (MB/sec)
Write - 8k random 20161 158 Write - 8k random 23905 187 Write - 8k random 25604 200
Read - 8k random 25501 199 Read - 8k random 27253 213 Read - 8k random 27293 213
Write - 64k sequential 8939 559 Write - 64k sequential 11905 744 Write - 64k sequential 14077 880
Read - 64k sequential 17317 1082 Read - 64k sequential 21320 1333 Read - 64k sequential 21731 13

regards,

Saurabh

  • Quincy561

    1281 Posts

    510

    1

    Posted October 1st, 2015 07:00

    Also if the tests are run separately,  you should allow time after a write test for the writes to destage before starting another test.

  • Quincy561

    1281 Posts

    510

    0

    Posted October 1st, 2015 07:00

    Is every test run separately? Did you run each one multiple times to see if you always get the same results?  Did you pre-write these devices?

    Not sure what impact the offset or the partition type has.

    Do you have some pre-set expectations for these tests or success criteria?