aic7xxx / AHA2940 worries... anyone?
Edward Welbon
welbon at bga.com
Sun Aug 10 14:34:18 PDT 1997
This discussion on Linux-kernel moved to this list as it seems to involve
AHA2940UW and aic7xxx.c as a central thread. I have had persistent crashs
with my 2940UW's. Too bad, as the attached file shows, in my system the
2940UW are fast.
On Sat, 9 Aug 1997, Daniel M. Eischen wrote:
> How's your SCSI bus laid out (what devices on what cables, where is
> termination, etc) and what options do you have enabled in the driver?
I carefully read and followed the recomendations of the SCSI3 spec for
cabling and termination. My cables are not over 1.5 meters, I have no
very short segments between the connectors. I terminate at the end of the
chain with the drives builtin terminators. I also explicitly set the
termination in the 2940UW. I also have tried other known good (and high
dollar) cables and have gotten the same results as my common sense
inexpensive cables. It ain't my cables.
As for my setup, I have three adapters (Asus PCI-sc875 is the adapter du
jour) and each adapter has three IBM DORS32160UW drives. I have
configured all nine disks as a raid0 (several partitions). Using initial
ram disk support, I mount my root file system on this raid0.
Here is my mdadd and mdrun.
/sbin/mdadd /dev/md0 /dev/sdb2 /dev/sde2 /dev/sdh2 \
/dev/sdc2 /dev/sdf2 /dev/sdi2 \
/dev/sdd2 /dev/sdg2 /dev/sdj2
/sbin/mdrun -p0 -c64k /dev/md0
With the BusLogic BT-958 and the Asus CPI-SC875, I have no problems; so
far I have run over 128 consecutive passes of Bonnie on files of 1GB or
larger on both adapters and no problems. I have really pounded the Hades
out of my disks with the BT-958, I have not tested the Asus quite so well
but the 2940UWs die miserably within about 3 or fewer consecutive passes
of bonnie on files of 2G size. The Asus handles the 2G files with aplomb
(though I have only done about 16 passes, it takes a while and I get bored
easily). Then the 2940 dies, a power cycle is required to restart.
Since two other adapters work perfectly, I am unconvinced that it is
my cables or my disks. Following are the non-commented configuration
details for my scsi drivers, I included all that I have used (and build
modules for) as a point of comparison.
CONFIG_SCSI_AHA1542=m
CONFIG_SCSI_AIC7XXX=m
CONFIG_AIC7XXX_TAGGED_QUEUEING=y
CONFIG_AIC7XXX_CMDS_PER_LUN=16
CONFIG_AIC7XXX_PAGE_ENABLE=y
CONFIG_AIC7XXX_RESET_DELAY=15
CONFIG_SCSI_BUSLOGIC=m
CONFIG_SCSI_OMIT_FLASHPOINT=y
CONFIG_SCSI_NCR53C8XX=m
CONFIG_SCSI_NCR53C8XX_TAGGED_QUEUE=y
CONFIG_SCSI_NCR53C8XX_MAX_TAGS=12
CONFIG_SCSI_NCR53C8XX_SYNC=20
CONFIG_SCSI_NCR53C8XX_SYMBIOS_COMPAT=y
I boot from the 1542 scsi and have no IDE or other boot media whatsoever.
Note that the scsi drivers above are all modules. Interesting eh?
/* One annoyance, Because /var is on the md0 raid, when I get a fatal
oops, I don't get anything in syslog. I once had /var/log on a different
file system though it didn't help, I still got no oops messages for scsi
driver barfs. I get oops messages for other things, I think that not
getting scsi oops messages is a consequence of the root file system mount
point. */
Ed Welbon; welbon at bga.com;
"We look for things"
-------------- next part --------------
I never intended to make this public so it may read too well.
The columns are essentially the same as those from Bonnie. The maximum
and minimum are not included in the averages so the first columns which
is the number of Bonnie passes is decreased by 2. The adaptec test was
run with the PCI "memory write with invalidate enable" control set (bit
4 of the PCI command register) set. It made very little difference,
I now use the defualt. In all cases the PCI latency is set to the max
of 248 on all master capable devices. The following is the raid set up:
/sbin/mdadd /dev/md0 /dev/sdb2 /dev/sde2 /dev/sdh2 \
/dev/sdc2 /dev/sdf2 /dev/sdi2 \
/dev/sdd2 /dev/sdg2 /dev/sdj2
/sbin/mdrun -p0 -c64k /dev/md0
Note that this setup causes sequential accesses to be interleaved across
both drives and controllers. I wrote the command out so that it would
be apparent (once it was pointed out 8-). It was a major help for me
(I dunno, about 15% I think?). It helps the most when raid is reading
on the outer cylinders.
Uh, lets see, the system is an ORION/82450 based SuperMicro P6DOF
dual Pentium Pro with the processors running at 80Mhz bus and 240
processor. The nine drives are IBM DORS32160UW's. System memory
is 128M. The adaptec is tested with a 2.1.42 kernel, the others with
a 2.1.48 kernel. The same Bonnie binary was used for all (with the
original 8Kbyte block reads).
adaptec aha2940uw results
tarantula:~/Bonnie# cat stats.K2.1.42.SCBF.M240.B80.3x2940UW.9xDORS.May8.wwk.md0.248.1024
AVERAGE 30.0 1024.0 10675.3 97.0 44533.7 84.3 8651.1 32.2 10944.4 77.3 16992.2 28.8 141.2 2.4
SCALED 30.0 1024.0 11009.2 100.0 52827.6 100.0 26883.6 100.0 14161.4 100.0 58932.6 100.0 5785.0 100.0
MAXIMUMS 30.0 1024.0 10734.0 97.5 45410.0 86.3 8752.0 33.2 11009.0 79.3 17312.0 32.2 142.9 2.9
MINIMUMS 30.0 1024.0 10575.0 95.8 41434.0 78.9 8522.0 28.6 10869.0 71.6 16714.0 23.9 138.2 1.8
SIGMA 30.0 0.0 30.5 0.3 654.5 1.2 42.4 0.8 25.8 1.7 133.8 1.2 1.1 0.2
SPREADS 30.0 0.0 5.2 5.6 6.1 6.2 5.4 5.6 5.4 4.6 4.5 7.0 4.4 4.6
COV 30.0 0.0% 0.3% 0.3% 1.5% 1.4% 0.5% 2.5% 0.2% 2.1% 0.8% 4.1% 0.8% 9.8%
ASUS PCI-sc875 results
tarantula:~/Bonnie# cat stats.K2.1.48.SCBF.M240.B80.3xPCI875.9xDORS.8k.md0.248.1024
AVERAGE 3.0 1024.0 10671.0 97.4 42643.7 83.1 8488.3 31.0 10893.0 76.9 16459.0 24.5 130.5 2.5
SCALED 3.0 1024.0 10952.1 100.0 51295.5 100.0 27411.2 100.0 14171.3 100.0 67088.3 100.0 5289.2 100.0
MAXIMUMS 3.0 1024.0 10686.0 97.6 43215.0 84.2 8543.0 31.3 10944.0 78.3 16607.0 25.5 131.1 3.3
MINIMUMS 3.0 1024.0 10636.0 97.3 40523.0 82.2 8456.0 30.1 10868.0 74.5 16368.0 22.9 128.8 2.0
SIGMA 3.0 0.0 11.3 0.0 161.6 0.4 10.9 0.0 8.5 1.1 38.6 0.6 0.5 0.4
SPREADS 3.0 0.0 4.4 6.4 16.7 4.4 8.0 25.5 8.9 3.4 6.2 4.4 4.3 3.5
COV 3.0 0.0% 0.1% 0.0% 0.4% 0.5% 0.1% 0.2% 0.1% 1.4% 0.2% 2.4% 0.4% 14.9%
BusLogic BT-958 results
tarantula:~/Bonnie# cat stats.K2.1.46.SCBF.M240.B80.3xBT-958.9xDORS.8k.md0.248.1024
AVERAGE 94.0 1024.0 10605.3 97.3 33767.2 77.4 8407.9 30.2 10834.6 76.0 16322.6 24.8 138.8 2.3
SCALED 94.0 1024.0 10903.6 100.0 43607.1 100.0 27853.6 100.0 14255.3 100.0 65816.9 100.0 5920.2 100.0
MAXIMUMS 94.0 1024.0 10681.0 98.0 34151.0 79.2 8520.0 31.5 10919.0 77.8 16536.0 27.1 144.0 3.3
MINIMUMS 94.0 1024.0 10450.0 95.8 32548.0 75.2 8215.0 28.6 6116.0 42.7 13919.0 21.5 133.5 1.9
SIGMA 94.0 0.0 53.2 0.4 289.5 0.8 41.5 0.7 32.3 1.2 97.9 0.8 2.5 0.3
SPREADS 94.0 0.0 4.3 5.4 5.5 4.7 7.4 4.1 148.9 29.3 26.7 6.9 4.2 5.0
COV 94.0 0.0% 0.5% 0.4% 0.9% 1.1% 0.5% 2.3% 0.3% 1.6% 0.6% 3.3% 1.8% 11.9%
Ed Welbon welbon at bga.com
Mission-Statement: "We look for things".
More information about the aic7xxx
mailing list