One reason I was setting up another iSCSI target to create a zpool
was to do some comparison of iscsi vs nfs, and ufs vs zfs.

So I have two pretty pathetic machines. Running S10U3 is
a SunBlade 1500 (1Ghz US-IIIi); the targets are on an
Ultra 60 running snv_54 with some old disk packs I scrounged
together. So these no way constitute benchmarks, but I
was interested in seeing what worked well and what worked
badly.

What I've tested is writing a large file using mkfile, and
using unzip to unpack an archive containing lots of small
files. Both are just write tests.

The configurations tested were - local ufs on the internal
disk in the SunBlade 1500; a ufs filesystem over iSCSI;
a zfs filesystem over iSCSI; nfs to the same zpool that is
backing the iSCSI targets; and nfs to a UFS filesystem
exported by the Ultra 60

100m mkfile:

local ufs - 2.758/2.708/2.416
iscsi ufs - 13.570/12.910/11.692
iscsi zfs - 1.322/1.647/0.809
nfs / zfs - 13.390/13.523/12.415
nfs / ufs - 9.624/9.080/9.100

Now, it hasn't made it to physical storage in under a second
(that would be impressive given that it's only running over
a 100M network), so the zfs case must include local caching.
Apart from caching effects, the 100M network is clearly coming
into play.

unzip of apache ant:

local ufs - 1.984/1.987/1.780
iscsi ufs - 5.497/5.304/4.578
iscsi zfs - 1.903/1.482/1.711
nfs / zfs - 1:32.679
nfs / ufs - 1:12.940

Ouch! I expected the nfs case to do badly (lots of small
files, with COMMIT for each one).

I'm very impressed with the numbers that iSCSI is giving me.

--
-Peter Tribble
http://www.petertribble.co.uk/ - http://ptribble.blogspot.com/
_______________________________________________
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss

Reply via email to