[Top] [All Lists]

Re: XFS fs size limit?

To: Seth Mos <seth@xxxxxxxx>
Subject: Re: XFS fs size limit?
From: Simon Matter <simon.matter@xxxxxxxxxxxxxxxx>
Date: Fri, 03 Aug 2001 08:01:33 +0200
>received: from mobile.sauter-bc.com (unknown []) by basel1.sauter-bc.com (Postfix) with ESMTP id CAA7157306; Fri, 3 Aug 2001 08:01:33 +0200 (CEST)
Cc: yocum@xxxxxxxx, xfs-list <linux-xfs@xxxxxxxxxxx>
Organization: Sauter AG, Basel
References: <200108021734.f72HYDo07108@jen.americas.sgi.com> <>
Sender: owner-linux-xfs@xxxxxxxxxxx
Seth Mos schrieb:
> At 13:05 2-8-2001 -0500, yocum@xxxxxxxx wrote:
> >Steve,
> >
> >I assume you mean software RAID5, right?  I've got a RAID50 array: 2 8-port
> >3ware cards in hw raid5, then those two devices combined software raid0.  I
> >didn't bother to make the log on a separate device, but I could, if need be.
> That is silly. Just make a big raid0 of it. If for some reason one of the
> 3ware cards takes a dump you WILL LOSE ALL YOUR DATA. I have seen this
> mistake (if it is) before at www.tweakers.net where they configured the 4
> disk into a raid 01 instead of 10. If you want redundancy make a raid0 on
> the 3ware disk and a raid1 of these 2 3ware cards.

I don't agree with this. Or maybe I do. If a 3ware card fails then the
raid0 over both cards will not start up and no changes are made to any
of the two raid5 volumes. When replacing a dead 3ware card it should be
able to run the existing disks and your raid0 will be okay with no data
lost. If the RAID controller can not import an existing array, then you
have lost your data (and should dump the controller to /dev/null).


> That is make 2 raid1 devices and then a raid0. If something happens to
> either one of the raid1 array logical devices (not disk but software
> failure) everything will be lost. Which was exactly what happend. They raid
> controller (Ami megaraid) got upset after one disk died and they lost there
> complete Mysql Database.
> > > this kernel revision is to use an external log device with raid5. You
> > > must also make sure to use an inode size larger than the default with
> > > a filesystem this big (mkfs ... -i size=512 ...). Without getting to
> > > technical, this will stop your inode numbers from overflowing 32 bits.
> > >
> > > >
> > > > Now there's rumors on l-k that the FS size might be limited by the
> > > > signedness of something in the block layer, effectively limiting us
> > to 1TB.
> > >
> > > We appear to have people working successfully beyond the 1Tbyte boundary,
> > > the 1Tb limit may be specific to some device drivers which are not as
> > > cleanly coded.
> >
> >
> >I haven't been terribly happy with the 3ware drivers/cards in general, so
> >we'll see.  FWIW, I tried upgrading to their latest driver on an smp system
> >and simply insmod-ing the newly compiled driver tickles the NMI watchdog
> >which dumps me directly into kdb.  Fun stuff.  :-/
> Which is exactly the reason to avoid the above mentioned configuration. If
> one card dies you have just lost all your data.
> Cheers
> --
> Seth
> Every program has two purposes one for which
> it was written and another for which it wasn't
> I use the last kind.

<Prev in Thread] Current Thread [Next in Thread>