LIPPKE@UTDALVM1.BITNET (David Lippke) (07/09/88)
We've recently put some effort into benchmarking various combinations of
TCP/IP software and ethernet controllers on IBM mainframes running VM.
The results of this testing have now been analyzed to death and we're
preparing for a second series of tests later this year.
Most of the performance problems of the current software/hardware in this
area are related to a long latency from packet reception to the processing
of that packet by the software (i.e., this latency tends to be much
greater than the transmit-commit to network latency).
I'm currently trying to quantify the effect that cutting one latency or
another will have on TCP performance. I think I may be homing in on
some decent conclusions and I plan on creating some live tests,
but it would be much easier to try out my theories if I had access
to some software which would simulate two interconnected TCPs.
I'm not after anything too complex; something that will simulate a single
established connection and let me adjust various parameters would be a
useful starting point.
Anyone have something even remotely applicable?
Kind Regards,
David Lippke
The University of Texas at Dallas
214-690-2632haas@utah-gr.UUCP (Walt Haas) (07/12/88)
In article <8807100545.AA19765@ucbvax.Berkeley.EDU> LIPPKE@UTDALVM1.BITNET (David Lippke) writes: >We've recently put some effort into benchmarking various combinations of > TCP/IP software and ethernet controllers on IBM mainframes running VM. > The results of this testing have now been analyzed to death... Could you be good enough to post your results? Thanks in advance -- Walt Haas haas@cs.utah.edu utah-cs!haas