value of the Beowulf cluster market

Mike Davis jmdavis at mail2.vcu.edu
Wed Jun 26 00:23:38 EDT 2002


I have to agree here, the value issue is exceedingly subjective even when you look at the products of turn-key
cluster vendors.

I currently manage two cluster of 48 and 42 dual proc nodes respectively. I also consult on a couple of smaller
clusters.

The main difference between these installation and others is that we made a decision several years ago to
implement  beowulf's for general purpose scientific computing. One of the two clusters is owned by VCU's CSBC
(Center for the of Biological Complexity) and reserves 1/4 of it's processors for general purpose computing. The
other is owned by Academic Technologies and is a resource available to researchers university wide. These
machines have let us commoditize our scientific computing by removing scalable and serial number crunching apps
from propietary hardware (SGI,SUN,AIX) and run more of these jobs on far less expensive machines. To provide an
idea of the efficency, we have more than doubled processor hours of grant funded research in the past year.

Compared to the work they support, the cost of the cost of these machines is relatively small even completely
built, pre-configured  and delivered. Support, training, and management are far more significant issues, IMO,
than hardware cost and maintenance.

Mike


"Robert G. Brown" wrote:

> On Tue, 25 Jun 2002, Dan Janies wrote:
>
> > Hello
> >
> > Does anyone know the current and or projected value of the Beowulf cluster
> > market?
> >
> > This may have been a recent thread but vigorous web/archive surfing didn't
> > produce.
>
> That's because the answer is probably "no".  To begin to build an
> answer, you'd have to define your terms pretty carefully.  I go to my
> local computer store and buy eight systems, a fast ethernet switch, add
> Linux, and put them together into a beowulf in my attic.  Have I
> participated in the "Beowulf Cluster Market" or just the generalized
> COTS PC market?  A rather large segment (in my unfounded opinion, a
> solid majority) of the beowulf clusters that exist are of this sort.
>
> There is a turnkey cluster market that is served by companies and
> individuals that do the buying, install a suitable Linux distribution,
> assemble the cluster for you, possibly accompanied by value-added
> software or custom features.  I don't know that anybody has ever done a
> formal analysis of the market -- one would have to figure out all the
> companies that do this and get annual reports for the last couple of
> years.
>
> As far as projected value is concerned, this is a double crapshoot --
> the turnkey market you MIGHT be able to estimate, but this is likely the
> tip of a large iceberg (even though there is some impressive ice out
> there).  The generalized homebrew market (including homemade rackmount,
> shelfmount, distributed/NOW/COW, and so forth) I wouldn't know how to
> begin to estimate.  Many clusters are very small (in my house I have a
> half-dozen nodes).  There are also a LOT of midsized clusters, and they
> add up.  In our department we have 68 nodes in six "distinct" small
> clusters (most of them dual processors) that are on a common network and
> CAN be used all together.
>
> The room is shared with another 30-40 nodes from the stats department.
> The math department also has a midsized cluster (maybe 64 nodes?).  Then
> there are chemistry clusters, biology clusters, genetics clusters -- all
> on one campus and in a couple of dozen research groups.  Very few of
> them were produced by turnkey vendors.
>
> So what's the "value" of this market to turnkey vendors?  I have no
> idea.  At a guess, only 1/5 or so of its "total" value, since beowulf
> clusters are by design easy to assemble out of commodity off the shelf
> parts.
>
>    rgb
>
> Robert G. Brown                        http://www.phy.duke.edu/~rgb/
> Duke University Dept. of Physics, Box 90305
> Durham, N.C. 27708-0305
> Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu
>
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf

_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list