[Beowulf] why we need cheap, open learning clusters

Lux, Jim (337C) james.p.lux at jpl.nasa.gov
Sun May 12 11:46:52 PDT 2013

On 5/12/13 11:29 AM, "Geoffrey Jacobs" <gdjacobs at gmail.com> wrote:

>On 05/12/2013 01:12 PM, Lux, Jim (337C) wrote:
>> Well.. On the Beowulf list the (super) goes without saying.. Is there
>> other kind worth considering?<grin>
>> (is a cluster of Arduinos a Beowulf? It's commodity hardware running
>> source software using commodity non-proprietary interconnects. I think
>> But I don't think it's HPC... So it would be a "personal cluster
>> or a "personal Beowulf" but probably not a "personal supercomputer")
>What do you call it when you must use multiple microcontrollers to
>achieve AD or DA conversion performance (in terms of sampling rate or
>number of channels) not possible with a single controller? Put another
>way, can a SCADA network be a Beowulf? One particular example that comes
>to mind is the EPICS system at CERN.

Interestingly, that's why I have those Arduino Ethernet widgets.

I think we could agree it's probably not HPC (although, if the sample rate
and bit widths are big enough, it might be)

I would think that it's not a Beowulf (or cluster computer) because the
overall purpose is not computation, but something else.  It's a cluster
DAQ system; and I leave it to others to find a suitable historical figure
with which to identify it.  Scylla perhaps?

"a monster with four eyes, six long necks equipped with grisly heads,
each of which contained three rows of sharp teeth. Her body consisted of
 twelve tentacle-like legs and a cat's tail while four to six dog-heads
ringed her waist."

Doesn't that describe the generally heterogenous nature of a data
acquisition system? It's even somewhat hierarchical with 6 heads/necks on
one body, and each head has 3 acquisition systems (or, perhaps that's just
triple modular redundancy?)

Argus, of the 1000 eyes, might be another possibility.

RGB defines a Beowulf (to distinguish it from,say a COW or NOW)

* The nodes are dedicated to the beowulf and serve no other

* The network(s) on which the nodes reside is(are) dedicated to the
beowulf and serve(s) no other purpose.

* The nodes are MMCOTS computers.  An essential part of the beowulf
definition (that distinguishes it from, for example a vendor-produced
massively parallel processor - MPP - system) is that its compute nodes
are mass produced commodities, readily available ``off the shelf'', and
hence relatively inexpensive.

* The network is also a MMCOTS entity (if not actually ``mass
market'' - some beowulf networks are sold pretty much only to beowulf
builders), at least to the extent that it must integrate with MCOTS
computers and hence must interconnect through a standard (e.g.  PCI)
bus.  Again, this is primarily to differentiate it from vendor-produced
MPP systems where the network and CPUs are custom-integrated at very
high cost.

* The nodes all run open source software.

* The resulting cluster is used for High Performance
Computing (HPC, also called ``parallel supercomputing'' and other names).

LittleFE and my ArduWulf (or is it a BeoDuino, or TeensyWulf) meet all
these criteria  (well, maybe not the HP part, but computing).  Your SCADA
example misses on a few of the criteria. (most likely just the last).


More information about the Beowulf mailing list