Coda File System

Re: [kind of offtopic] Re: CODA Scalability

From: Nick Andrew <lists_at_nick-andrew.net>
Date: Fri, 23 Aug 2002 19:14:38 +1000
On Thu, Aug 22, 2002 at 05:42:33PM +0200, Ivan Popov wrote:
> nbd on multiple machines, serving the disk space to a "filesystem server",
> that was going to run raid5 over the network "disks" and then put an
> encrypted filesystem on it, thus protecting the data integrity and privacy
> from corrupted disk space servers.
> [...]
> The project showed that raid is not suitable for "disks" that
>  - can silently contain wrong data
>  - can disappear and then come online again, without
>    manual intervention to "replace" them
>  - can become arbitrarily slow (depending on load and network)

Yes, I realised after reading your post, NBD for disks is a
silly idea (for a production system at least). When the disk
goes offline (and it will), then a big chunk of the filesystem 
is gone.

Also re your point about containing wrong data, RAID doesn't
protect at all against data errors on the disks, it merely
ensures that if one (or more) disks goes away for good, that
there is enough data remaining to reconstruct the original
data.

An example is, if you have two block devices in a RAID-1
arrangement and you deliberately write a different sector to
only one of the devices, then RAID will never notice this.
RAID clean or unclean state depends entirely on whether the
device was cleanly shutdown, not at all on the data on the
disks.

I will be looking forward to Serial ATA (www.serialata.com ?),
where drives plug into the host using a tiny cable which is
thinner and longer than existing IDE cables, making it easier
to put more disks in one housing.

Nick.
Received on 2002-08-23 05:20:09