Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

Subject: Re: [OMPI users] Theoretical Differences: Cluster Computing, Distributed Computing, Grid Computing
From: Aurelien Bouteiller (bouteill_at_[hidden])
Date: 2007-12-19 11:07:56


I have quite different definitions than Jeff.

Distributed computing is encompassing all the "parallel computing"
models, including clusters, grids, master-slave, shared memory...
Everything that basically implies using several collaborating
processes to solve a problem (whatever collaborating means, network,
shared memory, RPC, data dependencies... ).

Cluster computing simply refers simply to computations that occurs on
a cluster type machine. A cluster is -usually- a distributed memory
computer based on commodity hardware (but counter examples exist). It
may include commodity network (like giga ethernet) or more specific
nics (like myrinet, infiniband, quadrics and so on). As said Jeff,
usually the best way to use such a machine is to have a kind of
tightly coupled application.

Grid computing refers to gathering several clusters (and sometimes
large databases and scientific instruments like telescopes that
generate data) and use them altogether. Compared to a cluster or a
supercomputer, this introduces several issues related to password
administration, user domains, firewall bypass, several different
scheduler collaborating, and quite slow network between the sites.

Le 19 déc. 07 à 09:30, Jeff Squyres a écrit :

> It's a pretty hazy difference; I don't think there are formal
> definitions for each.
>
> Cluster computing *usually* (but not always) implies a bit more
> tightly coupled set of computing: the app communicates and coordinates
> between itself more than your standard manager/worker computation
> model. Distributed computing *usually* focuses on the manager/worker
> model -- send some work to anyone who asks for it and then eventually
> get some results back from them (or not -- then you have to send the
> same work out to someone else).
>
> But you can certainly use the manager/worker model in cluster
> computing, too. Not all cluster computing is tightly coupled.
>
> Probably the easiest distinction is that "cluster computing" occurs on
> a cluster; distributed computing [usually] occurs in a larger-than-a-
> SAN area.
>
>
> On Dec 19, 2007, at 1:10 AM, Amit Kumar Saha wrote:
>
>> Hi all!
>>
>> I am a bit hazy about the differences between Distributed Computing
>> and Cluster Computing.
>>
>> This document here
>> http://www.ibm.com/developerworks/grid/library/gr-heritage/ is a nice
>> read on Grid Computing and it also has few lines about important
>> differences from Cluster Computing.
>>
>> Hints will be duly appreciated!
>>
>> Thanks,
>> Amit
>> --
>> Amit Kumar Saha
>> Writer, Programmer, Researcher
>> http://amitsaha.in.googlepages.com
>> http://amitksaha.blogspot.com
>> _______________________________________________
>> users mailing list
>> users_at_[hidden]
>> http://www.open-mpi.org/mailman/listinfo.cgi/users
>
>
> --
> Jeff Squyres
> Cisco Systems
> _______________________________________________
> users mailing list
> users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/users