RAID controller question...

[Kragen]

On Thu, 10 Sep 1998, Josip Loncaric wrote:
> Greg Lindahl wrote:
> > As usual, generalizations lead you astray. If you want to hook 10+
> > drives to a machine and are seek limited, UIDE won't do it for you,
> > even if your sustained bandwidth is much lower than 20 MB/s. Think
> > database. Sure, that means you spend more money.
> 
> I thought that the idea of the original Beowulf was to use many
> processors with a disk or two per CPU, with the goal of achieving bigher
> aggregate bandwidth than the alternative configuration of many disks per
> CPU.

The Vesta distributed filesystem from IBM did this.  I'm curious if
anyone has done this with Linux machines -- Vesta was for an SP.

Is it possible to use the new "access raw disk across a network"
facility in Linux 2.1 (does it exist or is it a rumor?) along with the
md drivers to begin to simulate this?

Such an approach would also provide some amount of redundancy --
although the machine running the md driver would appear to be an SPOF,
there could be a backup machine prepared to run the md driver as soon
as the current master failed, and start fscking.

This would have no SPOF, except that fscking a huge disk takes
forever.  (Pray for journaling in ext2!)

Vesta was superior in this regard, because it did not have a single
point of failure (for performance reasons) -- the clients had to figure
out what machines to talk to to get their files.

Kragen (who hasn't used Vesta, the md driver, two-tailed SCSI disks,
	or even Linux 2.1)

-- 
<kragen@pobox.com>       Kragen Sitaker     <http://www.pobox.com/~kragen/>
I don't do .INI, .BAT, .DLL or .SYS files. I don't assign apps to files. I 
don't configure peripherals or networks before using them. I have a computer 
to do all that. I have a Macintosh, not a hobby. -- Fritz Anderson