Hi Tomo, as said it was a simple test. Yes I changed all tgt parameters to match iet. I'll try your test. Another thing of course is, both initiator and target ran on the same machine. So, I'll use a different machine for the target and initiator and run the test you did. Unfortunately, they're only coupled on 100 MBit ethernet. Let you know the results. Albert FUJITA Tomonori wrote: > On Mon, 26 Nov 2007 09:53:38 +0100 > Albert Pauw <apauw at inter.nl.net> wrote: > >> Just reminded myself to compare apples with apples, not pears ;-) >> >> So I ran the tests again, both type of targets (iet, tgt) with the same >> settings >> (MaxBurstLength, DataSegmentLenghts, Digests). >> >> I see a six fold difference in performance, I don't see iet starting low >> (twice the performance of tgt) > > So IET is still faster? > > If so, very wired. I just run open-iscsi and IET/tgt with two boxes > connected via a GbE switch with the same iSCSI parameter > > sens:/home/fujita# ./ltp-full-20061222/testcases/kernel/io/disktest/disktest -PT -T30 -h1 -K8 -B32768 -pL -ID /dev/sdd > > - tgt > > | 2007/11/26-17:39:24 | STAT | 2184 | v1.2.8 | /dev/sdd | Read throughput: 53706752.0B/s (51.22MB/s), IOPS 1647.0/s. > > - IET > > | 2007/11/26-17:44:21 | STAT | 2248 | v1.2.8 | /dev/sdd | Read throughput: 42860544.0B/s (40.88MB/s), IOPS 1316.0/s. > > > BTW, you need to change several parameter with tgt since the tgt > default paramaters are different from the IET paramaters. > _______________________________________________ > Stgt-devel mailing list > Stgt-devel at lists.berlios.de > https://lists.berlios.de/mailman/listinfo/stgt-devel > |