* Re: swap on raid 1
2003-10-14 13:36 ` Hermann Himmelbauer
@ 2003-10-14 14:03 ` Gordon Henderson
2003-10-14 14:30 ` Thomas Steudten
1 sibling, 0 replies; 5+ messages in thread
From: Gordon Henderson @ 2003-10-14 14:03 UTC (permalink / raw)
To: linux-raid
On Tue, 14 Oct 2003, Hermann Himmelbauer wrote:
> On Tuesday 14 October 2003 13:34, skopel@english.fsu.edu wrote:
> > I have been following the discussion of swap on raid 1 and I just want to
> > clarify something... I have my swap on /dev/md1, which is a raid1 device
> > consisting of /dev/hda2 and /dev/hdc2.
> > If one of the drives fails (either hda or hdc), I think the swap will be
> > duplicated on the other and the box will continue merrily on, right?
> > thanks
>
> Basically, that's what I hope.
>
> Anyway - what exactly happens when a PATA harddisk dies? Will this only lead
> to logentries in the syslog, or will this crash the machine due to hardware
> reasons?
Who knows. I do know that if you plug the cable into an IDE drive the
wrong way up, the processor won't start, so I guess it's possible that a
failing IDE drive can take out the entire PC if you are unlucky.
> Probably there would also be a difference if you would do RAID on
> the same channel (e.g. hda, hdb) as a mulfunction of one device could jam the
> whole IDE channel.
This is generally bad. Don't do it. See archives and HowTos for more
reasons not to. I've had it personally happen to me (ie. a failing drive,
not part of a RAID set that time, caused the other drive on the cable to
be unreachable)
> The same thing can apply to SCSI, too - I once experienced this myself
> when a fautly SCSI-CDROM jammed the whole SCSI-bus and crashed my
> machine.
>
> I hope that SATA will be a lot less critical, it should support
> hotplugging (at least with SATA 1.1, it's a pity that it's not supported
> with 1.0 right away) and there is only one device per channel.
In theory you can hot-plug (and unplug) an IDE drive, with various
hardware caddys and incantations of hdparm, etc. to spin it down and
remove it from the kernel, but I don't recomend it. The last time I had a
real drive failure which Software RAID coped very well with, I scheduled a
maintenance reboot and changed the drive then.
Gordon
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: swap on raid 1
2003-10-14 13:36 ` Hermann Himmelbauer
2003-10-14 14:03 ` Gordon Henderson
@ 2003-10-14 14:30 ` Thomas Steudten
1 sibling, 0 replies; 5+ messages in thread
From: Thomas Steudten @ 2003-10-14 14:30 UTC (permalink / raw)
To: Hermann Himmelbauer, linux-raid
USB, SCSI, PATA, SATA or whatever, this are buses, and only one device
keeps a line down or high from this bus, every device can´t work correct
any more.
So to be fault resistence, you need two controllers with only one device
connected to it, or two controllers connected to every one device
(dual channel). So if one controller fails, the device can talk over
the other one. And if the device fails, you can HOPE that it won´t
keep the two (SCSI) interfaces in a invalid state.
On the other hand with two controllers, you have only one systemboard or cpu
which can fail too.
With raid you can catch device errors like, bad-blocks, disk error, but
not that the device/ disk keeps the bus.. (HW Error in the interface circuit).
Tom
Hermann Himmelbauer wrote:
>
> Anyway - what exactly happens when a PATA harddisk dies? Will this only lead
> to logentries in the syslog, or will this crash the machine due to hardware
> reasons? Probably there would also be a difference if you would do RAID on
> the same channel (e.g. hda, hdb) as a mulfunction of one device could jam the
> whole IDE channel.
>
> The same thing can apply to SCSI, too - I once experienced this myself when a
> fautly SCSI-CDROM jammed the whole SCSI-bus and crashed my machine.
>
> I hope that SATA will be a lot less critical, it should support hotplugging
> (at least with SATA 1.1, it's a pity that it's not supported with 1.0 right
> away) and there is only one device per channel.
>
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 5+ messages in thread