Subject: Re: Building a huge file server
To: Markus W Kilbinger <mk@kilbi.de>
From: David Laight <david@l8s.co.uk>
List: current-users
Date: 07/05/2006 22:30:52
On Wed, Jul 05, 2006 at 01:45:33PM +0200, Markus W Kilbinger wrote:
> Hi!
> 
> Im planning to build a huge file server. My idea was to put about 8
> big IDE or SATA disks (500 - 750 GB each) into a PC/i386 based system
> and create a big raid(4) on it (RAID level 5, I thought).
> 
> I guess I have to use ffs2 to create partitions larger than 1 (2?)
> tera bytes. Does this work sufficiently, yet?

Both FFSv1 and FFSv2 can handle multi-TB filesystems.
The restriction for FFSv1 is 2^32 (or maybe 2^31) filesystem fragments.

Since FFSv2 uses 64bit fragment numbers, the size of the on-disk inodes
and indirect blocks doubles - which increases the wasted space.

FFSv2 does 'lazy initialisation' of on-disk inodes which speeds up newfs
(and might improve fsck slightly).

As mentioned by others, the netbsd disklabel can describe disks above 2TB,
but you could put a single large filesystem onto the raw device.
However the last person that tried that found filesystem corruptions that
I've not seen anyone fix.

Testing disks > 2TB is beyond most people at home :-)

	David

-- 
David Laight: david@l8s.co.uk