[GE users] Best cluster management software and OS

Brady Catherman bradyc at uidaho.edu
Sat Nov 19 23:14:19 GMT 2005

Very well said =) Im sorry for dragging the thread down there.

Remember though. When building a cluster. Do not, I repeat, do NOT  
rely on a single person for all of your answers. I have seen a good  
half a dozen clusters built completely wrong because the person  
offering advice was either used to a single type of cluster, or in  
the "throw money at the problem" mind set.

If your programs require heavy communication/file access avoid a NFS  
mounted root. You only burden your network with more load. If you are  
running jobs like Paup, that run for weeks and don't do communication  
or heavy file access then NFS root is an awesome solution.

If you are running zero communication jobs you might want to look  
into Condor or other "use existing hardware" solutions. If you decide  
that a cluster is still in your best interest go with a 100M switch..  
They are cheap and you won't tell the difference..

Cluster in a box solutions often leave much to be desired and tend to  
miss the mark when it comes to applications you want to run. (Thats  
not to say that they don't work at all, often if you only run BLAST  
then a BLAST oriented cluster in a box solution is a great idea!)

I think there is only one thing that is consistent for every single  
cluster out there. Do not let it overheat. Spend money on a good AC  
system and make sure you watch the temps.

Anyways, look for a good solution to YOUR cluster needs. (I think  
this is one of the reasons why I love clusters so much is that there  
is so much room for creativity in design and such. And throwing  
hardware at the problem often just makes the problem worse =)

On Nov 19, 2005, at 9:03 AM, Hai Pham wrote:

> Sebastien is right, we've strayed way off the original poster's
> question.   Whether  you use distribution A or Y isn't really that
> important at the end of the day.  The most important question you
> should ask yourself is what do you want to do with the cluster once
> it's done.  What software are you planning to run?  How much in house
> expertise do you have to build the system if you plan to do it
> yourself?  How much time do you have to spend on building it?  How
> much money do you have for software?
> A good complete answer to this question goes far further than just
> Redhat is best or Gentoo is the fastest.  And it's way beyond whatever
> "advice" that I might write about in this post.  For starters, the
> fact that the original poster asked what open source OS to use, not
> what Linux distribution to use makes me think that he/she might not
> have as much experience with clusters and operating systems in general
> as we're assuming.
> First of all, if you are looking for open source operating systems for
> clusters, Linux is really the only game in town.  Educate yourself --
> posting to a website and asking isn't going to give you a magic
> answer.  Go to Google and search for "beowulf clusters" and then read.
>  Yes, unfortunately, RTFM.  There are lots of good stuff out there
> about how to build a cluster.  The more you know about what's
> involved, the better you'll be to make an educated decision about how
> to build your environment.
> Next, I think the biggest mistake that a lot of people make when it
> comes to software is to ask "what can I do to get the most performance
> out of the hardware".  I think that's the wrong question to start with
> -- very few people need the absolute best performance from their
> hardware.  The question that they should really ask is "what can I do
> to get the most productivity out of the hardware".  Notice that they
> are not really the same question.  It's like the difference between an
> F1 car and a Toyota Corolla.  If you need to get groceries, an F1 car
> might get you there faster, but you're going to have to spend a lot
> more time and resources into it to get it going at all.  In the
> meantime, you could have gone to the grocery store, cooked dinner and
> put your kids to sleep in the amount of time that you it took you to
> get your F1 car started.
> This brings up the point about what Linux distribution to use.
> Everyone has their favourite Linux distro, each distro has it's pro
> and con.  As someone who has been using Linux since version 0.11 (yes,
> that was when Linus himself built the "distribution"), I have used
> just about every major Linux distribution over the years.  None of
> them are perfect, so the choice of which one to use will largely
> depend on what you are used to and what you are planning to install on
> top of it.  The pros and cons of each distro is a totally different
> topic and everyone you ask will give you a different opinion about
> their favourite.
> The question you should be asking yourself in this case, is what
> software am I planning to run on this cluster?  For example, we
> decided not to try and roll our own clustering environment, but to buy
> off the shelf to save time.  After some investigation, I decided that
> Bioteam's iNquiry gave us a starting place that most closely resembles
> what we wanted to do with our cluster.  iNquiry works best with SUSE
> 9.1, so that's the distro that we went with.
> The next thing to consider is what cluster management system to use.
> If one or more of your applications require SGE, then there's your
> answer.  If you decide to go with an off the shelf clustering
> solution, it might already come with a management system built-in.
> Also, if you know someone with experience with one of the available
> systems, don't discount their knowledge.  They are an extremely
> valuable asset for the times when you get stuck.  Likewise, don't
> forget about the community of users that are using a particular
> clustering solution or management system.  When something breaks and
> you need help, having a big community of users to draw on can save you
> all sorts of time and headaches.  Personally, I tried a number of
> different management systems before deciding that I liked SGE the most
> -- mainly because it has a very active user base and it has most of
> the features that I find useful, not because I thought it was the best
> or because it gave me the best performance.
> Do your due dilligence, the time you spend educating yourself now will
> save you tons of time later!
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net

To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net

More information about the gridengine-users mailing list