Messages in this thread | | | Date | Thu, 28 Mar 1996 13:08:50 +0100 | From | Egbert Eich <> | Subject | Re: SCSI bug |
| |
Andrew C. Esh writes: > sirdir wrote: > > > > Hi > > I encoutered a 'bug' in the SCSI subsystem of Linux. I have a harddrive > > that spins down now and then to do some sort of recalibration (silly > > thing, I know... but otherwise it's fast and big) .. And yes, if I say it > ...
This discussion is on the wrong list. You should direct it to the SCSI list.
> > > > anyways, i hope somebody can fix this problem... i love linux, but with > > this harddrive it's somewhat unusable... > > The drive is unreliable anyway. It's not the fault of the OS. If you > get it repaired, let me know what they did to it. I need to have mine > fixed too. (Is it possible you have the drive in power-saver mode?) >
The drive is not broken. Some of the low-priced IBM-drives do this, they have to go into a maintenace mode every 72 hours to recalibrate, therfore they spin down completely for 1-2 minutes (I don't recall exactly). At least it's smart enough to do some statistics about disk usage and tries to avoid times of high disk access. IBM makes no secret out of this, it's right in the specs. IBM does not recommend these drives for high performance system like UNIX machines which are up all the time or RAIDs etc. . They are mainly built for the low price DOG market, no problems are expected there, since who ever got a WINDOG system running for 72 hours in a row without a breakdown? But on the other hand, who can resist this low price?
> > BTW: SCSI drives are *not* supposed to spin down by themselves without > being instructed to do so by the OS. This would be catastrophic. >
Tell IBM about it. They tell you to spend a little more money, then you can get (almost) the same drive from them which does not show this behaviour.
I don't know, what the drive does with incoming requests. If it just accepts them and completes them, when it's 'clean-up' is done, it might help to increase the timeout for SCSI requests (from 1 s. - I believe - to 180 s.). Then all requests which do a disk request will lock. If it they dont even accept the request then it's more involved to make the kernel handle this more gracefully. To say anything about it I'd have to see the exact error messages.
-- Egbert Eich.
Egbert.Eich@Physik.TH-Darmstadt.DE
| |