[CentOS] Desktop Supercomputer

Sun Jul 18 02:15:25 UTC 2010
Les Mikesell <lesmikesell at gmail.com>

Rajagopal Swaminathan wrote:
> Greetings,
> 
> Thanks Les for your reply. (I dont top-post normally, but this was an emergency)
> 
> On 7/17/10, Les Mikesell <lesmikesell at gmail.com> wrote:
> 
>> Have you looked at Ubuntu's setup?  I don't think it deals with GPU's but it
>> might be an easier starting point than building from scratch, and if
>> anything
>> outgrows your resources it can move to Amazon's ec2.
> 
> The point is, why should we not have and use our own resources?
> 
>> http://www.ubuntu.com/cloud/private

If you had read the link you would have seen that it does use your own resources 
- that's why it is called a private cloud.  Or did you mean that CentOS should 
have an equivalent to match the Ubuntu configuration?  You probably could, but 
it would be a lot more work.

>> But, I think you need to look at storage and compute facilities differently
>> in a
>> cloud model.  Storage needs to be HA and redundant.  Computing needs to be
>> able
>> to fail and be replaced.
> 
> 1. I am not able to understand  your above statement clearly. (see my
> earlier replies on this thread)

Distributed (super) computing normally involves techniques that break the work 
up into independent units.  If a node doesn't complete a unit it can be given to 
a different node.


> 2. I believe (and have experienced including and mother and father),
> that nothing and/or everything is (ir)replacable.
> 
> 4. I am not confused (See point 1)

Storage, on the other hand, needs to be reliable.  The hadoop project is an 
approach at distribution, but the interface is hard to deal with.

-- 
   Les Mikesell
    lesmikesell at gmail.com