[4488] in linux-scsi channel archive
Re: AIC7XXX + MD RAID0 at Alpha are sure-fire receipe for a hang...
daemon@ATHENA.MIT.EDU (Matti Aarnio)
Fri Jul 31 21:04:50 1998
In-Reply-To: <19980731223517Z92201-389+1@mea.tmt.tele.fi> from Matti Aarnio at "Aug 1, 98 01:35:09 am"
To: matti.aarnio@sonera.fi (Matti Aarnio)
Date: Sat, 1 Aug 1998 04:01:19 +0300 (EEST)
Cc: linux-raid@vger.rutgers.edu, linux-scsi@vger.rutgers.edu
From: Matti Aarnio <matti.aarnio@sonera.fi>
(3:30 AM, I had to catch the last commuter train (of the night) to get
to home from office; the first train of the morning goes in
two hours, though.. Now I can tell how my problem was solved.)
> Hello,
Hello again!
This Linux-hackerdom synkronicity is occasionally unbelievable :-)
I reported problems with Linux-2.1.112 on Alpha (SMP commented away),
and using stock AIC7XXX driver and stock MD/RAID0 stuff.
Ingo Molnar announced new RAID utility and kernel patch set just
seconds after my letter went thru, and I was already patching his
new bits into 2.1.112 when he wrote mentioning of them.
I ALSO installed AIC7XXX 5.1.0pre5 driver here at the same time,
and got the beast to boot just fine. (A full kernel compile for
me takes 15 minutes, but I have a 'sluggish' uniprocessor Alpha
system..)
Now storing files into that RAID0 storage works reliably :-)
A beauty-wart is that 'df' doesn't know how to handle these
'large' system sizes, but that is not RAID* problem at all.
$ df
Filesystem 1024-blocks Used Available Capacity Mounted on
/dev/sda3 1717459 1095269 533431 67% /
/dev/sdb1 4044375 2468111 1367019 64% /home
/dev/md0 16870424 1138510 14853322 7% /usr/src
I can definitely recommend AIC7XXX 5.1.0pre5 for an AHA 2740UW
controller. I also can recommend the new RAID kernel modules,
which in fact do behave slightly differently from stock 2.1.112
raid modules:
(old)
raid0 : Allocating 16 bytes for hash.
raid0 : Allocating 16 bytes for hash.
(new)
raid0 : Allocating 16 bytes for hash.
raid0 : Allocating 1072 bytes for hash.
Ingo can tell, what the difference might have been caused by..
Oh yes, something I didn't already tell Ingo:
In raidtools-0.50* source module raid_io.c does have
horribly UGLY way of adding llseek() syscall for the
utility's uses.
In GLIBC 2 there is llseek() built in. (And at Alpha -- well,
sizeof(long long) == sizeof(long) :-) ) Do see how Ted Ts'o
handles it in his ext2 utilities toolbox autoconfiguration.
/Matti Aarnio <matti.aarnio@sonera.fi>
(4:00 AM, time to hit the pillow...)
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.rutgers.edu
