[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: multicore management system



Hello,

Douglas A. Tutty wrote:
On Fri, Dec 28, 2007 at 10:00:02AM +0800, Jerome BENOIT wrote:
(On this list, we either intersperse comments or bottom post)
Douglas A. Tutty wrote:
On Thu, Dec 27, 2007 at 06:57:20AM +0800, Jerome BENOIT wrote:
I am looking for a management system for a multicore system (2 and 4):
namely an alternative to `at' tools.
Any suggestion is welcome.
More information needed.

Are you talking about sending pending jobs to free cores?  Doesn't the
kernel scheduler just do that itself?

What problems have you had with the 'at' tools that you need to address?
I made some progress since my post:
I am not really looking for an alternative to `at',
but rather something similar `Torque':

www.clusterresources.com/pages/products/torque-resource-manager.php

May be this tools is too powerful for a single computer:
hence my request.

There's a huge difference between managing a cluster and managing
cores/CPUs on one system.  The biggest, is that one system runs one
kernel whereas the cluster runs one for each box.
AIUI, the linux kernel will use one core until it gets to 80% used (20%
idle or so) before it powers-up and begins to use a second core.  Also,
most apps aren't written to be able to use multiple CPUs.
Single-threaded apps only use one CPU. What is it about 'Torque' that you find you need on your box?


I have a long list of jobs that I want to launch on my box.
On a one core box, I can make a for loop in order to submit then
in sequence. On a multi-core box, this approach does not work anymore.
Hence my intention to use Torque: here I envisage my multi-core
box as a cluster where each core is a node. Note that you can configure
LAM in such a way.

Jerome


Doug.



--
Jerome BENOIT
jgmbenoit_at_mailsnare_dot_net


Reply to: