From: "Kent Nilsen" <[EMAIL PROTECTED]>
Date: Wed, 8 Sep 1999 13:48:11 +0200
I've also got some questions concerning the Mylex 960:
I've got a JBOD box with 5 Seagate Elite23 disks, which is connected to a
Mylex DAC960 PG controller (channel 0). On channel 2 (internal) I've got 5
Barracuda 18Gb disks. The machine is a PIII450 with 384Mb RAM, a
DAC960PG, Supermicro Dual PII mainboard with onboard 789x SCSI. Running
Mandrake 6.0, kernel 2.2.7 with the latest patches available
The Elites used to be mounted on a Sun computer, and caused no trouble
there. The barracudas were mounted in a JBOD cabinet on a Sun, and caused
no trouble. However, when I put the Barracudas in the server, I had to disable
Tagged command queuing to make it work (else it said "bad tag received from"
whatever disk was ID 0.). I also connected the JBOD cabinet with the Elites to
channel 0 on the DAC960. I got that initialized, and formatted it, mounted it,
and finally started using it. After a while I began to get some weird messages:
DAC960#0:Physical Drive 0:10 Error Log: Sense Key=6, ASC=29, ASCQ=00
------------"---------------------------------------------: Information = 00000000
00000000
And that message is repeated for each disk (0:11, 0:12 etc). I don't have any
disks on the lower addresses (because the disks were set this way when I took
over). /proc/rd/status says "OK", /proc/rd/c0/current_status reports no
problems.
The above messages indicate that a bus reset has occurred. Starting with the
2.2.4 version of the DAC960 driver a counter of bus resets is kept for each
drive rather than logging errors every time. If a command to a disk drive
is not completed within 6 seconds, the DAC960 times it out, resets, and
reissues the command. This is not a problem unless it occurs frequently.
The "bad tag returned from drive" tends to be a catchall error that generally
indicates buggy firmware in the disk drive. Needing to disable tagged command
queuing also indicates a likely problem. Have you checked with Mylex to be
sure you have the latest version of DAC960PG firmware and whether there are
known problems with the firmware in the Seagates?
It takes me 3-4 minutes to mount it, and these error messages pops up all the
time.
The last part is the System drive setup... I tried to add 5 more Elite23
disks on Channel 1, but when I plugged the disks in without setting them up,
the DAC960 1: Did not set up any of the previously configured system drives.
2: Did not find the last 23Gb disk (only 4 of 5 disks, and this box just came
from another server that's been running on the disks for almost a year.). 3:
When I tried to delete one system drive (the other 23Gb Elites) to make a new
bigger one, it says it can only delete the last created system drive. That's
my 80Gb RAID5 disk with lots of important data! This isn't a big problem,
I've got backup, but taping in and out takes time. And does that limit really
increase security or whatever it's there for?
The above seems strange. The DAC960 should have still been happy with the
configuration of the existing drives, leaving the new ones as unconfigured, and
allowing you to build a new drive group and logical drive out of them.
All in all, is the DAC960 a very "strict" card that won't work with cables
that work with most others? My cable length is about 1.50 meters, with a 0.6
meter cable and 0.9 meter approx inside the JBOD. My terminators are active,
and I have no 68->50 pin converters on the chain.
First, 5 drives per channel may well be violating the UltraSCSI specification.
The DAC960PG might well be quite picky in this regard. Have you tried setting
the transfer speed down from 40MB/sec to 20MB/sec to see if avoiding the faster
signalling helps?
The newer Ultra2/LVD controllers seem to be a lot better in regard to cabling
issues.
Leonard
