Poor IO performance EVA4000 hpux

[Antonio Gonzalez_4]

Hi all.
We are having poor perfomance working with an EVA4000, mainly for write operations. We have 18MB/s for READ and 120MB/s for WRITE operations.
READ: 17,85MB/s
sprod:/ARCHIVING#time dd if=RFI04280.020040.VMPCUSTOM.ARCHIVE
of=/dev/null bs=8192 count=131072
6399+1 records in
6399+1 records out
real 2.8
user 0.0
sys 0.2
WRITE: 120MB/s
sprod:/ARCHIVING#time dd if=/dev/zero of=prueba.txt bs=8192 count=131072
131072+0 records in
131072+0 records out
real 8.5
user 0.1
sys 4.8
kernel parameters we have:
Some parameters:
dbc_max_pct 8 - 8
dbc_min_pct 5 - 3
bufpages 0 - (NBUF*2)
nbuf 0 - 0
We have 2 simliar HBA with this info:
sprod:/#ioscan -fnC td0
sprod:/#ls -al /dev/td0
crw-rw-rw- 1 bin bin 49 0x000000 Aug 29 00:19 /dev/td0
sprod:/#tdutil /dev/td0

Vendor ID is = 0x00103c
Device ID is = 0x001029
XL2 Chip Revision No is = 2.3
PCI Sub-system Vendor ID is = 0x00103c
PCI Sub-system ID is = 0x00128c
Topology = PTTOPT_FABRIC
Link Speed = 2Gb
Local N_Port_id is = 0x010000
N_Port Node World Wide Name = 0x50060b0000246c8d
N_Port Port World Wide Name = 0x50060b0000246c8c
Driver state = ONLINE
Hardware Path is = 3/0/6/1/0
Number of Assisted IOs = 865926384
Number of Active Login Sessions = 1
Dino Present on Card = NO
Maximum Frame Size = 2048
Driver Version = @(#) libtd.a HP Fibre Channel Tachyon TL/TS/XL2 Driver B.11.11.12 PATCH_11.11 (PHSS_31326) /ux/kern/kisu/TL/src/common/wsio/td_glue.c: Sep 5 2005, 10:14:40

I enclosed also the result of
STM > TOOLS > UTILITY > RUN > LOGTOOL > FILE > RAW SUMMARY

We are working with informix in raw device

sar -b
00:00:00 bread/s lread/s %rcache bwrit/s lwrit/s %wcache pread/s pwrit/s
....
16:10:00 0 429 100 303 611 51 2372 640
16:20:00 0 479 100 365 731 50 2643 528
16:30:00 0 467 100 423 840 50 2491 497
16:40:01 1 637 100 382 787 52 2381 589
16:50:01 0 497 100 292 586 50 2250 594

Average 3 4921 100 99 211 53 2252 647

sar -d

Average c0t8d0 3.98 0.50 6 49 0.00 8.53
Average c1t8d0 1.20 0.50 2 21 0.00 7.30
Average c11t0d1 99.60 0.50 479 3484 0.00 7.49
Average c13t0d2 75.10 0.50 155 2230 0.00 8.17
Average c11t0d3 99.00 0.50 498 4887 0.00 5.84
Average c13t0d4 90.24 0.50 258 2626 0.00 8.39
Average c11t0d5 86.25 0.50 226 1673 0.00 9.96
Average c13t0d6 97.41 0.50 168 870 0.00 10.26
Average c11t0d7 43.23 0.50 45 223 0.00 12.60
Average c13t1d0 72.11 0.50 72 338 0.00 14.47
Average c11t1d1 97.41 0.50 280 1145 0.00 7.74
Average c13t1d2 47.21 0.50 56 293 0.00 10.62
Average c13t1d6 40.84 0.50 28 111 0.00 17.65
Average c11t1d7 34.06 0.50 26 105 0.00 15.22
Average c13t2d0 56.97 0.50 36 154 0.00 15.37
Average c11t2d1 48.21 0.50 24 97 0.00 23.64
Average c13t2d2 50.00 0.50 38 153 0.00 15.62
Average c11t2d3 94.02 0.50 80 1246 0.00 11.55
Average c13t2d4 91.04 0.50 123 1903 0.00 7.36
Average c11t2d5 2.99 0.50 120 1855 0.00 0.25
Average c13t2d6 90.24 0.50 192 2974 0.00 4.98
Average c11t2d7 2.39 0.50 94 1466 0.00 0.26

We saw many 'blocked' processes into vmstat:

$ vmstat -S 1
procs memory page faults cpu
r b w avm free si so pi po fr de sr in sy cs us sy id
2 22 0 2841181 52682 111056 111098 4 4 0 0 66 6703 38944 9726 30 5 65
2 22 0 2841181 52627 0 0 2 0 0 0 0 7553 73615 20391 38 16 47
2 22 0 2841181 52627 0 0 1 0 0 0 0 7578 73903 21166 41 14 46
2 22 0 2841181 52627 5 5 1 0 0 0 0 7584 74552 21806 39 13 49
2 22 0 2841181 52627 0 0 0 0 0 0 0 7685 70071 20438 22 7 71

Do you have a disk bottleneck?
Thanks.

[Antonio Gonzalez_4]

We have checked READ performance improves not using PV´s. It seems it´s better to leave EVA doing the data distritution alone across the disk and not to use PV from the operating system to distribute over the EVA:

vgsap50/rlvfrag1_52 (1 unique PV)
vgsap50/lvbtab_52 (lvol distributed over 5 PVs)

sprod:/#time dd if=/dev/vgsap50/rlvfrag1_52 of=/dev/null bs=8192 count=5000
5000+0 records in
5000+0 records out

real 0.9
user 0.0
sys 0.1
sprod:/#time dd if=/dev/vgsap50/rlvbtab_52 of=/dev/null bs=8192 count=5000
5000+0 records in
5000+0 records out

real 17.8
user 0.0
sys 0.1

Antonio.