Re: Still looking for thin clients?

[gentry@xxxxxxxx]

At 10:48 PM -0900 10/30/03, Joshua Kugler wrote:
>We have a group working on that right now.  For more information e-mail James
>Gentry at gentry@arsc.edu
>
>Hope that helps.
>
>- Joshua
>--
>Joshua Kugler
>Assistant Systems Administrator
>UAF Department of Math and Sciences
>UAF LUG President


Hello Mike,

We currently have a group of people interested in building a cluster 
and I'm the "head geek" for the project.  Nice to meet you.  :-)

Our next LUG meeting is this Thursday (11/06/2003).  I am going to 
arrange a meeting of the cluster group during the weekend of 11/08 - 
11/09.  Dayne has apparently found some exciting software that could 
make building a cluster much easier.

Your idea about utilizing wireless as an interconnect is interesting 
but would probably result in very poor performance.  Although I 
suppose it depends on what is being computed on the cluster.  In most 
cases, we are going to want the fastest interconnect possible, with 
the lowest latency possible.  This would rule out wireless.

That being said...  there is no reason why we can't have 2 or more 
interconnects on the same cluster.  It would be interesting to see 
what sort of performance we could get over wireless.  I have been 
thinking about a way to create a job scheduler such that when a user 
places a job in the queue, they decide which interconnect they want 
to use and how many nodes.  If the user has an 'embarrassingly 
parallel' problem, then they won't need a fast interconnect.  On the 
other hand, if they have a job that will require a lot of 
intercommunication between the nodes, they will want to select the 
faster interconnect.

Another thing I want to discuss is funding for a 'real' cluster.  I 
was recently at the IEEE Vis 2003 conference in Seattle and let me 
tell you...  clusters are THE hot topic right now.  Graphics 
clusters, strictly computational clusters, you name it and if it has 
"cluster" in the title, people are interested.  I would like to 
propose to the group that we build a small Linux cluster from old 
donated equipment and use this as a training exercise.  While we are 
doing this, I want to explore funding possibilities for building a 
large computational cluster.  I believe we can get decent funding or 
donations to build a good sized cluster.  Obviously, there are many 
issues to be discussed surrounding this idea.  It will take time and 
a lot of work.  I am planning on being in Fairbanks for the next few 
years and involved with clusters at my work.  I would like to 
integrate into LUG some of the cool stuff happening in the 
visualization and computational sciences world.

So...  if you are interested, look me up this Thursday at the 
meeting.  We are going to be getting the ball rolling on this cluster 
project.

-- 

James Gentry III
Visualization Programmer Analyst
Arctic Region Supercomputing Center
University of Alaska Fairbanks
(907) 474-1104
gentry@arsc.edu