[Stgt-devel] ibmvio is slow

[Bastian Blank]

On Tue, Jun 19, 2007 at 07:18:38PM +0200, Bastian Blank wrote:
> But it is slow. I can write at most 15MiB/s to the disks, regardless of
> the count of the initiators and disks.

I did another round of tests with 2.6.22 and a corresponsing patchset
and an uptodate tgtd with disabled controller lun (it breaks my setup
which needs the boot disk on lun 0).

Writing on an exported device shows weird behaviour. iostat shows many
write _and_ read requests, even if the initiator only emits write
requests:

| Device:         rrqm/s   wrqm/s   r/s   w/s   rsec/s   wsec/s avgrq-sz avgqu-sz   await  svctm  %util
| sdb            3733.90  4177.60 525.80 72.70 34138.40 34002.40   113.85    19.94   33.32   1.38  82.76

If the initiator emits read requests, it is okay:

| Device:         rrqm/s   wrqm/s   r/s   w/s   rsec/s   wsec/s avgrq-sz avgqu-sz   await  svctm  %util
| sdb           13798.50     0.40 1955.31  0.20 126057.72     4.81    64.47     1.18    0.60   0.43  83.21

There is only lvm between it. This looks like some sort of read-ahead
for the data, which should not happen for write requests. I tried to dig
a little bit in the block code but was not able to found the cause.

Bonnie shows the following:
| Version  1.03       ------Sequential Output------ --Sequential Input- --Random-
|                     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
| Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
| test.track.rz.un 2G           21914  10 15207   4           59795   6 305.8   0

Bastian

-- 
Emotions are alien to me.  I'm a scientist.
		-- Spock, "This Side of Paradise", stardate 3417.3