Questions about the Personal IRIS

[Vernon Schryver]

In article <8812260910.aa25159 at SMOKE.BRL.MIL>, XBR2D96D at DDATHD21.BITNET (Knobi der Rechnerschrat) writes:
> 2) During excessive use of a Network-Test-Program called 'dog'
>    (listening Reinhard ????), we discovered that people inside
>    the Personal IRIS side of the pipe were really handicaped compared
>    to a 4D/70G. This brought us to the conclusion that the Personal-
>    Iris Ethernet adapter is slower that the adapters for the G and GT.
>    Is this observation correct?
> 
> Martin Knoblauch
> West-Germany
> BITNET: <XBR2D96D at DDATHD21>

The UDP/IP/ethernet in the 4D20 is the fastest we currently ship.  It
speaks directly to memory, rather than making the CPU stroke it over the
VME or other bus.  Perhaps one of the graphics or CPU-&-cache-hardware
experts could say if the performance difference is there.  Doesn't dog do a
lot of floating point?  Does your 4D20 have a floating point chip?

You might want to measure relative performance using the other interactive
network stress test, 'arena'.  In all of this, be careful.  You should have
seen what arena did to old minicomputers manufactured by a large eastern
company when arena ran unthrottled, at > 50 pkts/sec/machine.

Vernon Schryver
Silicon Graphics
vjs at sgi.com