<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2//EN">
<HTML>
<HEAD>
<META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1">
<META NAME="Generator" CONTENT="MS Exchange Server version 6.5.7638.1">
<TITLE>RE: [Beowulf] Compute Node OS on Local Disk vs. Ram Disk</TITLE>
</HEAD>
<BODY>
<!-- Converted from text/plain format -->
<BR>
<P><FONT SIZE=2>Ram disks, definately. ;-)<BR>
<BR>
Afraid I'm still going with diskless nodes. You save<BR>
<BR>
1. some money on the disks themselves<BR>
2. more money on solving disk failures<BR>
3. yet more money on cooling<BR>
<BR>
This may be specially important in a high-density rack situation, where if you get the disks out of the way each box can get more ventilators up front as well as in the rear.<BR>
<BR>
On the other hand, Windows Vista has had its uses(!), such as driving RAM prices down as demand expands ...<BR>
<BR>
Cheers,<BR>
-Alan<BR>
<BR>
<BR>
<BR>
-----Original Message-----<BR>
From: beowulf-bounces@beowulf.org on behalf of Jon Forrest<BR>
Sent: Mon 9/29/2008 6:44 AM<BR>
To: Beowulf Mailing List<BR>
Subject: [Beowulf] Compute Node OS on Local Disk vs. Ram Disk<BR>
<BR>
There are two philosophies on where a compute node's<BR>
OS and basic utilities should be located:<BR>
<BR>
1) On a local harddrive<BR>
<BR>
2) On a RAM disk<BR>
<BR>
I'd like to start a discussion on the positives<BR>
and negatives of each approach. I'll throw out<BR>
a few.<BR>
<BR>
Both approaches require that a compute node "distribution"<BR>
be maintained on the frontend machine. In both cases<BR>
it's important to remember to make any changes to this<BR>
distribution rather than just using "pdsh" or "tentakel"<BR>
to dynamically modify a compute node. This is so that the<BR>
next time the compute node boots, it gets the uptodate<BR>
distribution. Although the mechanism for maintaining<BR>
the distribution varies in either approach, I consider<BR>
this a push since one mechanism isn't inherently better<BR>
than the other.<BR>
<BR>
Assuming the actual OS image is the same in both cases,<BR>
#2 clearly requires more memory than #1. There are actually<BR>
two approaches to #2 - a) where only the OS and other stuff<BR>
necessary to boot the system are kept in memory and everything<BR>
else is in an NFS-mounted file system, and b) where the whole OS<BR>
installation is kept in memory. Depending on which approach<BR>
is taken, the RAM-based installations can take hundreds of MB<BR>
more than a local harddrive installation. However, on a modern<BR>
multicore compute node this might just be a few percent of the<BR>
total RAM on the node.<BR>
<BR>
Long ago not installing a local harddrive saved a considerable<BR>
about of money but this isn't true anymore. Systems that need<BR>
to page (or swap) will require a harddrive anyway since paging<BR>
over the network isn't fast enough so very few compute nodes<BR>
will be running diskless.<BR>
<BR>
Approach #2 requires much less time when a node is installed,<BR>
and a little less time when a node is booted.<BR>
<BR>
What are some of your favorite issues, positive or negative, with<BR>
each approach?<BR>
<BR>
Cordially,<BR>
--<BR>
Jon Forrest<BR>
Research Computing Support<BR>
College of Chemistry<BR>
173 Tan Hall<BR>
University of California Berkeley<BR>
Berkeley, CA<BR>
94720-1460<BR>
510-643-1032<BR>
jlforrest@berkeley.edu<BR>
_______________________________________________<BR>
Beowulf mailing list, Beowulf@beowulf.org<BR>
To change your subscription (digest mode or unsubscribe) visit <A HREF="http://www.beowulf.org/mailman/listinfo/beowulf">http://www.beowulf.org/mailman/listinfo/beowulf</A><BR>
<BR>
</FONT>
</P>
</BODY>
</HTML>