X-PMC-CI-e-mail-id: 11374
Hi,
Roland Boden wrote:
>I think you got the right idea!
>I enabled "allow disconnect" for the tape device in the Tekram
>controller and now it seems to work. I could read some tapes
>correctly.
Great.
>I now also understand why I couldn't do anything with the disk while
>trying to do something with the tape, even if the disk and the tape
>are connected to different scsi hosts (you wrote the scsi bus is being
>monopolized).
Hmm... I think there is something fishy going on.
Tape is on one scsi card, and the
disk is on the DIFFERENT card, aren't they?
Then my question would be why the non-disconnect-capable tape drive on
one scsi-bus interferes with the data transfer on a DIFFERENT scsi
bus?
I think it should NOT.
Is this a linux scsi system bug?
This reminds me of one thing.
I believe there was a discussion about
the loss of paralelism of processing in linux SCSI subsystem
several months ago.
I can't write a detailed description right now, but someone in Japan
did a rather comprehensive test of scsi disk performance and he did a
nice comparison of using striping and others using raid tools on
Linux. To my great surprise, his data suggests that using TWO (2)
SCSI controllers doesn't help. That is, using two disk controllers to
distribute the disk load on the separate bus does NOT improve
the performance of SCSI disk at all on linux. Mirroring, even using
dual CPU (!) doesn't show any perceptible performance difference when
you use one (1) or two (2) SCSI controllers.
He did various testing using the combination of 1-2 CPU, 1-2 SCSI
controller cards.
Doesn't this sound strange?
Does the tape problem have something in common with this obeservation
of the SCSI performance with one or two scsi controller cards?
Is there a serialization bottleneck in Linux SCSI system?
I will get to write a more description about his performance data,
hopefully next week (if Japan survives the worst nuclear accident,
that is. Check your local news NOW!).
--
Ishikawa, Chiaki [EMAIL PROTECTED] or
(family name, given name) [EMAIL PROTECTED]
Personal Media Corp. ** Remove .NoSpam at the end before use **
Shinagawa, Tokyo, Japan 142-0051
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to [EMAIL PROTECTED]
