montnaro@sprite.crd.ge.com (Skip Montanaro) (10/28/89)
Does anyone know if Cogent Research is shipping a product? If so, does anybody have any comments to offer on it? From what I recall they are/were working on a Transputer-based workstation. Each workstation could have a small number of processors, but communicated to other Cogent w/s's using fiber optic technology. Linda is supposed to be in the kernel (or even the foundation for the kernel's communication). As an aside, wasn't there a comp.sys.linda newsgroup? It's not in my .newsrc any more and GNUS can't find it. I wondered why it disappeared (if it isn't just a figment of my imagination). Thanks, -- Skip Montanaro (montanaro@crdgw1.ge.com)
segall@caip.rutgers.edu (Ed Segall) (10/31/89)
Cogent demo'd a product here at Rutgers recently. I believe they have
been shipping for about 6 months, but I'd check that figure to be
sure.
I was relatively pleased with the hardware. The fully configured
system targets what I consider to be a medium performance range (160
MIPS, 48 MFLOPS spec for current availability). They did mention that
they plan to turn up the processor clocks in the near future, giving a
higher peak MIPS/FLOPS rating. Not quite a parallel KILLER MICRO, but
certainly respectable.
I don't know how well that would scale up, however. Each Resource
Server can support up to 32 T800s, plus additional workstations, each
with dual T800s. There is a bus, so I doubt it's practical to put
more processors on each server. Certainly, using Linda would
facilitate interconnecting multiple Resource Servers, if you're
looking for total FLOPS, but I think the minumum useful grain size
would increase rapidly when you start doing that. Caveat - I have no
information on which to base that opinion - it's just a hunch.
I was much more impressed with the software. They built QIX (Unix
compatible) on top of Linda (actually, Kernel Linda). This meant that
once Kernel Linda was right, most Unix concurrency issues became Linda
concurrency issues, which is a very clean platform for concurrent
programming. Vollum confirmed, when I asked him, that building on top
of Kernel Linda made Unix development significantly faster and easier
than it would otherwise have been.
My impression was that for Linda development, this is an excellent
vehicle. For number-crunching that doesn't strain it's limits, it
sounds very nice. Beyond that, we'll have to wait and see what they
do with the hardware.
Charles Vollum, Chairman and Chief Technical Officer (503/690-1450,
vollumc@ogcadmin.OGC.EDU) would have more information.
--Ed Segall
--
uucp: {...}!rutgers!caip.rutgers.edu!segall
arpa: segall@caip.rutgers.edu