On Sun, Apr 12, 2009 at 8:09 PM, Bart Van Assche
<bart.vanass...@gmail.com> wrote:
> While running iSCSI performance tests I noticed that the performance
> for certain block sizes deviated significantly (more than ten times)
> from the performance for other block sizes, both larger and smaller.
> This surprised me.
>
> The test I ran was as follows:
> * A file of 1 GB residing on a tmpfs filesystem was exported via iSCSI
> target software. The test has been repeated with both SCST and STGT.
> * On the initiator system open-iscsi version 2.0.870 was used for
> performing reads and writes with dd via direct I/O. Read-ahead was set
> to zero.
> * Both systems were running kernel 2.6.29.1 in run level 3 (no X
> server) and the 1 GbE interfaces in the two systems were connected via
> a crossed cable. The MTU has been left to its default value, 1500
> bytes. Netperf reported a throughput of 600 Mbit/s = 75 MB/s for the
> TCP/IP stream test on this setup.
> * 128 MB of data has been transferred during each test.
> * Each measurement has been repeated three times.
> * All caches were flushed before each test.
> * The ratio of standard deviation to average was 2% or lower for all
> measurements.
> * The measurement result are as follows (transfer speeds in MB/s):
>
> Block   SCST    STGT    SCST    STGT
>  size  writing writing reading reading
> ------ ------- ------- ------- -------
>  64 MB  71.7    63.3    62.1    58.4
>  32 MB  71.9    63.4    61.7    58.1
>  16 MB  72.4    63.0    61.7    57.1
>   8 MB  72.7    63.3    61.7    56.9
>   4 MB  72.9    63.5    61.3    57.0
>   2 MB  72.8    59.5    60.3    56.9
>   1 MB  72.1    38.7    59.4    56.0
> 512 KB  67.3    21.4    58.0    54.4
> 256 KB  67.4    22.8    55.5    53.4
> 128 KB  60.9    22.6    53.3    51.7
>  64 KB  53.2    22.2    53.0    45.7
>  32 KB  48.9    21.6    40.0    40.0
>  16 KB  40.0    20.8     0.6     1.3
>   8 KB  20.0    19.9    19.9    20.0
>   4 KB   0.6     1.6    18.9    10.3
>
> All results look normal to me, except the write throughput for a block
> size of 4 KB and the read throughput for a block size of 16 KB.
>
> Regarding CPU load: during the 4 KB write test, the CPU load was 0.9
> on the initiator system and 0.1 on the target.

By this time I have repeated the above test on the same two servers
but this time using a DDR IB network using IPoIB in connected mode
instead of a GbE network (netperf reports 1375 MB/s). The results for
the IB network look normal to me. But it's still not clear to me why
for certain block sizes the number of IOPs was so low on the GbE
network.

 Block   SCST    STGT    SCST    STGT
 size  writing writing reading reading
------ ------- ------- ------- -------
 64 MB   307     251     658     439
 32 MB   460     258     630     431
 16 MB   580     258     598     414
  8 MB   585     262     566     413
  4 MB   665     273     613     428
  2 MB   737     240     645     476
  1 MB   727     172     631     457
512 KB   623     115     583     326
256 KB   619     120     559     399
128 KB   467     118     499     306
 64 KB   309     115     333     251
 32 KB   233     107     248     198
 16 KB   157      95     162     133
  8 KB    97      79      95      78
  4 KB    51      42      52      42

Bart.

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups 
"open-iscsi" group.
To post to this group, send email to open-iscsi@googlegroups.com
To unsubscribe from this group, send email to 
open-iscsi+unsubscr...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/open-iscsi
-~----------~----~----~----~------~----~------~--~---

Reply via email to