Re: Compute Farm
On Tue, 10 Feb 1998 bhmit1@mail.wm.edu wrote:
> On Tue, 10 Feb 1998, Tim Sailer wrote:
>
> > Stephen Carpenter wrote:
>
> > > hmmm just a slightly evil thought....
> > > anyone tried
> > > cat /dev/hda1 > /dev/hdb1
> > > assuming hda1 and hdb1 are similar partition sizes and types....
>
> > Tried it... doesnt work.
>
> The correct answer is "dd if=/dev/hda1 of=hdb1 bs=1024". This doesn't
> care what is one the drive, even win 95 (known by experience), and it will
> act like a perfect mirror.
>
> The proper thing to do is to nfs export everything on one system, make a
> boot disk that can mount the nfs drive, partition/mount the new drive, and
> edit any specific info (hostname and ip) all by a simple boot script.
> It's way above my head, but when set up right, would be almost as easy to
> do 100 as it is to do 5. I think this is what the titanic team did (in
> the lj article). I don't think they used redhat's easy method.
I used this (well, similar) method to install 58 machines in the computer
lab. They're all same, with 2.1G hard drives. I installed both linux and
windoze95 (well, it wasnt me, but a collegue ;)) on one machine (about
50%-50% for linux and w95). I made a boot disk, which sets up the network
with bootp, booted the installed machine off it, nfs mounted the server,
and then gzipped the whole disk (/dev/hda) with dd if=/dev/hda | gzip -c
>/mnt/diskimage.gz. Then I booted another machine with the bootdisk,
mounted the server and did a gunzip -c /mnt/diskimage | dd of=/dev/hda. I
run 4-10 of these simultaneously (had to start them at the same time, so
they dont slow down because of reading different parts of diskimage.gz). I
didn't need to edit any bootup scripts, everything is setup thru bootp. It
all went perfectly, no need for partitioning either, /dev/hda contains
that info too. Windows95 had some problems though, it cannot get its name
from the bootp server, and it had to install another network card driver
because its pnp serial number changed. With linux this was no problem, I
used a pnpdump/sed script on /etc/isapnp.conf before running isapnp.
So this installation went perfectly, the only problem is that it is slow.
58*480Megs (this was the diskimage.gz size) thru a 10Mb network ... well,
it took around 10 hours. Since then I thought of a better, faster way to
do this, I've already written the programs for it, just cant test it yet.
So the idea is: run a program on the already installed machine, which
opens /dev/hda and then udp broadcasts it to the network. Each client
catches the broadcasted packets, which has some information about block
number, block size and a crc, checks it, writes it to the disk, and marks
the block received. When the server finishes, it sends a block with length
0, then the clients can send querys to the server for the missed blocks.
I'll try these programs in a few weeks... I expect it to finish the
installation of the whole lab in less than an hour.
Greg
--
Madarasz Gergely gorgo@caesar.elte.hu gorgo@linux.rulez.org
It's practically impossible to look at a penguin and feel angry.
Egy pingvinre gyakorlatilag lehetetlen haragosan nezni.
HuLUG: http://www.cab.u-szeged.hu/local/linux/
--
TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to
debian-user-request@lists.debian.org .
Trouble? e-mail to templin@bucknell.edu .
Reply to: