Beowulf in a Box (fwd)

[Simon Thorpe]

Bob Glamm wrote

>First, if I recall my PCI spec correctly, it is possible to configure
>each PCI device with an address space.  It should be possible to give
>each StrongARM module (CPU + PCI interface + memory) its own
>distinct physical address space outside of the memory of the host
>machine.  In addition, this address space should probably be
>uncacheable as well to the host computer for a first iteration.

Yep. From what I understand, it will be possible to assign each StrongARM
its own private memory space (in which case the system can have as many of
the CPUs as you want within the host). Alternatively, you can have all the
daughterboard memory in the host's memory space. With 64 Megs per processor
this will allow 64 processors (i.e. 8 boards) in the same system all able
to access each others memory. Another possibility is to have (say) half the
memory as private, and the other half mapped into the hosts memory space (I
seem to remember that the 21285 bridge chip allows you to assign blocks of
16 Megs in various combinations). If people are interested, then its worth
looking at the 21285 documentation
http://developer.intel.com/design/bridge/dsc-21285.htm . It really is a
neat chip, you can program it to do all sorts of nice things - at least you
can if you work for Causality... ;-)

>Second, it is not necessary to run the entirety Linux on any of the StrongARM
>processors.  Doing so in this configuration (in any configuration
>IMHO ;) is a waste of resources.  In fact, why bother to put
>ethernet on board these PCI cards?  Well written (and scalable)
>MPI programs don't do a lot of communication; one single
>Gigabit ethernet card in the host should be plenty of network
>resource for quite a few StrongARMs.

Actually, my original idea when I spoke to Neil Carson and colleagues was
that with our neural network simulation code we really wouldn't need an
operating system at all - just the minimum for launching the program, and
picking up and sending messages. Later on, we came to the conclusion that
it would make sense to use up a few megs of RAM in order to have a NetBSD
or Linux kernel running. First, it would make it much easier to get going,
and second, it would mean that anybody interested in Beowulf type clusters
would be able to use it too... :-)


__________________
Simon Thorpe
Centre de Recherche Cerveau et Cognition
133, route de Narbonne
31062 Toulouse France
Tel 33 (0)5 62 17 28 03. Fax 33 (0)5 62 17 28 09
http://www.cerco.ups-tlse.fr/private/simon.html
__________________