Subject: Re: differential scsi controllers
To: None <root@garbled.net>
From: Berndt Josef Wulf <wulf@ping.net.au>
List: port-i386
Date: 07/02/1998 07:58:51
Tim Rightnour wrote
> 
> On 01-Jul-98 Thor Lancelot Simon spoke unto us all:
> #  Even if it worked, it wouldn't be worth using.  It's absolute idiocy --
> #  mirroring for the sake of being able to say you're mirroring.  It always
> #  reads from the first component of a mirror set, thus giving no performance
> #  benefit, and it doesn't handle errors *at all*, meaning that an error on
> #  either component will cause the halves of the mirror to become inconsistent,
> #  so that is very unlikely to provide redundancy, either.  It's Awful.  It
> #  should have been removed Long Ago.
> 
> I hate to say it, but AIX has a really nice software mirroring.  If the drive
> in question bites it, it does a few things:
> 
> A: It marks the partitions on the bad drive as being stale, and no longer
> reads/writes them if a previous read/write failed.
> 
> B: Noticing a stale mirror, allows you to break that mirror, (without affecting
> anything runnning on it)  Assuming you have a hot-swappable drive, you can then
> throw it in, and resync the mirror, and go on with life.
> 
> Even if we dont have a hot-swappable media, it would be nice to be able to
> break the mirror, and continue with life until the new drive arrives and we can
> shut down.  No applications need to die until you actually shut the machine
> off, and most importantly, *no data is lost*.
> 

I must agree here as I know out of first experience that Novel-Servers
around the globe use the mirroring technique in view of providing
redundancy against harddisk/controller failure.

After recovering from a harddisk crash last week by just replacing 
a faulty 9GB drive and resyncing the partitions, I have started to 
appreciate this technology and so do the network users. 

Maybe a worthwhile effort of fixing the current code to become more
efficient in disk mirroring instead of throwing it out altogether.

cheerio Berndt
-- 
Name	: Berndt Josef Wulf
E-Mail	: wulf@ping.net.au
Sysinfo	: DEC AXPpci33+, NetBSD-1.3