[8154] in linux-scsi channel archive

home help back first fref pref prev next nref lref last post

Re: recovery behaviour with 1 bad + 1 good drive (aic7xxx)

daemon@ATHENA.MIT.EDU (Ricky Beam)
Tue Feb 22 15:08:15 2000

Date:   Tue, 22 Feb 2000 14:17:55 -0500 (EST)
From: Ricky Beam <jfbeam@bluetopia.net>
To: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Matthias Andree <ma@dt.e-technik.uni-dortmund.de>,
	linux-scsi@vger.rutgers.edu
In-Reply-To: <E12NJke-0002HT-00@the-village.bc.nu>
Message-ID: <Pine.LNX.4.04.10002221404330.12259-100000@beaker>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII

On Tue, 22 Feb 2000, Alan Cox wrote:
>> Why, some RAID controllers are cheaper than Adaptec SCSI cards :-)
>
>A raid controller wont help you on a bus hang. I learned that the hard way
>last night when a MegaRAID got a bus hang and totalled my /usr/src in the
>process as I had it set to write back. 

First rule, never set a RAID card to write-back.  (esp. if you don't have
a BBU attached to it.)  Don't acknowledge writes until they are actually
on the physical disk(s).  I've had the same problem with a DPT controller
in a news server -- people would power it down or hit the reset button before
it was done flushing 128M of write behind data.  HOWEVER, the news server
was not designed for reliablity; it was designed for speed.

(I know, I'm a fanatic.)

This depends on the firmware of the controller.

In my case, it would appear something as simple as a SCSI bus parity error
can crash a Mylex RAID controller.  I don't know why the hell they won't
listen to me when I tell them it's crashing -- I even sent them the i960's
console data from boot to crash.  (I sent the board for repair.  Two months
later, I get it back with a new RTC battery on it !?!?  That, of course,
didn't fix a damned thing -- it crashed 17 min. after I put it back in the
system.  Needless to say, this REALLY pisses me off.)

(Note to the reader, the "Mylex Only" connector on the AccelRAID cards is
 actually a serial port connected to the i960.  19200-8-n-1.)

>Ext2fs isnt happy when the last 20Mb of so of writes,(not in the order it wanted
>them to happen either) go missing.

Indeed.  Add to that the effects of dcache and you have a recipe for
mke2fs :-)

--Ricky



-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.rutgers.edu

home help back first fref pref prev next nref lref last post