Is it possible to provide the iometer test configuration's that are used for testing the iscsi open-e performance. I am using iometer but not sure what configs to use to determine what the actual performance is.
Also are there any base line figures for what the Read and Write throughput be for a good spec open-e server
There can be many issues on why you would not be maxing the NIC’s or transfer speeds. IO Meter is just many to use and test with. I use DTR2000, I set up a batch file that looks like this e: being my drive to test. I averaged at 64MB when done. I have seen better then mine personal system that is because they buy the best components.
In our 2006 Test of Open-E iSCSI vs SanMelody Toms Hardware USA tests we have our documented speeds around 80MB/s. Try bonding NIC’s using 802.3ad (please set switch to accept these Bonding Modes) or test with balance ALB as well. Test with Jumbo Frames (Switch and host must be able to except these parameters). Does the Disk Controller have the capability to in-crease Cache write values. Are the drives fast? Check host systems NIC’s, overhead CPU usage. Was the RAID set up for large chunk size (highest setting)? When a lot of very small files, have to be copied, this leads to a big overhead and not fully using TCP/IP packages. The next reason is, that small files are mostly spread over the disks, and the Disks are permanently seeking for files. So many things have to be considered and there are so many variables to understand from the environment to the engineer’s knowledge. Are most of the files Reads or are they Writes, size of files (large or small)? Is this mostly Random access? The RAID may not be fast enough. There are also MS 2.02 Initiator documents to review and test to improve speeds provided by Microsoft.
Whizz Kid
Performance testing
Posting Permissions
Reply With Quote