Printable View
Hi All,
I'm new here, i met a problem, the iscsi performance is descendant heavily, i used software RAID, created an iscsi volume based on RAID5, and then connect the target with windows initiator, and run 1MB sequency write test with iometer, it has 100MB/s mostly when the test starting, but after the test ran 12 hours, the speed is 50MB/s, Is there anybody has met the problem?
Since this can be directly related to hardware, some more details are needed here.
Give us some specifics as to what you are testing, and what the settings are, in regards to your setup.
Hi Gr-R, Thank you for your reply, we just test the iscsi volume performance to see what it can be used for.
the hardware configuration is:
Supermicro motherboard X7DCX
4*2GB DDR2 physical memory, speed is 667 MHz (1.5 ns)
LSI SAS 3081E-R
Intel Dempsey/San Clemente/ICH9, Intel(R) Xeon(R) CPU E5405 @ 2.00GHz
2*1GB physical network adapter
Thanks again!
Here is the result of the iscsi volume performance with IOZONE on windows and linux.
Linux: the iscsi volume formated the volume as ext3 with default parameters
Writer Report
64 128 256 512 1024 2048 4096
4194304 90152 99316 90032 92065 81018 85747 86212
Re-writer Report
64 128 256 512 1024 2048 4096
4194304 86036 85099 90097 87432 83687 81529 80045
Reader Report
64 128 256 512 1024 2048 4096
4194304 94822 97099 96894 97612 60799 60636 60795
Re-reader Report
64 128 256 512 1024 2048 4096
4194304 92670 97945 96336 97591 60543 60611 60505
Random Read Report
64 128 256 512 1024 2048 4096
4194304 7269 12359 17589 27667 41623 56222 65754
Random Write Report
64 128 256 512 1024 2048 4096
4194304 8905 10600 19286 33684 42681 54249 56901
Windows 2k3: the iscsi volume formated as NTFS with default parameters
Writer Report
64 128 256 512 1024 2048 4096
4194304 8560 10600 12901 8223 6298 9974 6003
Re-writer Report
64 128 256 512 1024 2048 4096
4194304 31411 36873 43424 7099 6871 9749 5928
Reader Report
64 128 256 512 1024 2048 4096
4194304 52838 55762 54605 60310 69646 48959 56039
Re-reader Report
64 128 256 512 1024 2048 4096
4194304 59710 64809 59356 67408 70002 49021 56189
Random Read Report
64 128 256 512 1024 2048 4096
4194304 7068 8140 10155 18138 28435 33137 42907
Random Write Report
64 128 256 512 1024 2048 4096
4194304 8895 6726 9790 4423 7211 9649 5374
According to the result, it's more and more better random write performance when the written date size changed from 64KB to 4MB on linux. but on windows, the random write performance is so poor, always below 10MB. Is there any operation need to do?
What version are you running and is there Write Back option on the LUN? Is the Write cache enabled on the RAID controller?
make sure firmware and dss are current versions.
search the forum for this:
MaxOutstandingR2T - 8
MaxRecvDataSegmentLength - 262144
MaxXmitDataSegmentLength - 262144
and make the tweaks from the console.
write back can help too.
Hi Todd, Thanks for your reply, write back option on the LUN was enabled, and what's the meaning of the Write cache enabled on the RAID controller? I used the software RAID, not hardware RAID. thanks!
Hi Gr-R, it's seems that the parameters you pointed had incorrect value, i will try it again with you suggestion, thank you very much!
Hi Gr-R, I changed the configuration parameters, the performance did not improve. Is there any other reason?
What version of DSS you using?
Is hardware RAID an option ?
See this link on running IO tests.
http://blog.open-e.com/a-few-practic.../#comment-4743
or run BST5 (NTFS formatted) http://www.nu2.nu/download.php?sFile=bst514.zip
Sequential 1MB for reads and writes
Sequential 4MB for reads and writes
Hi Gr-R, There is no hardware RAID, we're using the software RAID,and the DSS version is 4622.
Update to 5087.Quote:
Originally Posted by daobang wang
Some updates to SCST that may help you.
For DSS 5087, Is there any other change except the SCST updated to 2.0.0.1? I'm afraid that will not help, because the SCST 2.0.0.1 changed as follows:
- Fixed critical issue with wrong INQUIRY page 0x80 (r3197)
- Possible crashes due to incorrect execution context estimations fixed
- Responsiveness under heavy load improved
But i will try it, thanks!
there is more than just the SCST update in the latest version.
go ahead and update, and see if it helps.
careful to read the release notes about VFD's and vmware for the update.
Hi Gr-R, please see the iozone test output for the same configuration for iscsi volume with windows client. It seems that the performance is no improve, thanks!
DSS 4622
Writer Report
64 128 256 512 1024 2048 4096
4194304 11534 18050 14847 7474 7808 9359 7804
Re-writer Report
64 128 256 512 1024 2048 4096
4194304 32558 36580 42025 7321 7794 9305 7698
Reader Report
64 128 256 512 1024 2048 4096
4194304 49648 50696 53375 58658 62220 45873 50432
Re-reader Report
64 128 256 512 1024 2048 4096
4194304 59039 58341 53921 67305 62955 45322 51352
Random Read Report
64 128 256 512 1024 2048 4096
4194304 7080 8071 10063 18171 26656 31538 39510
Random Write Report
64 128 256 512 1024 2048 4096
4194304 8988 8486 8412 5328 6972 8994 7583
DSS 5087
Writer Report
64 128 256 512 1024 2048 4096
4194304 9734 7995 8227 6551 6958 7051 8845
Re-writer Report
64 128 256 512 1024 2048 4096
4194304 36120 36559 37083 6525 6792 7146 8865
Reader Report
64 128 256 512 1024 2048 4096
4194304 51351 50456 51729 59453 61986 44904 50413
Re-reader Report
64 128 256 512 1024 2048 4096
4194304 59174 58035 59063 63760 63601 45156 49738
Random Read Report
64 128 256 512 1024 2048 4096
4194304 7147 7871 12205 16032 26345 34340 37549
Random Write Report
64 128 256 512 1024 2048 4096
4194304 10066 8143 7853 5529 6492 7005 8671