Comments

Thu Sep 1 17:50:49 UTC 1994

> Curtis,
>  
> We have done tests on two switches with various configurations - 10 
> sparcstaions, 3 7000's, AGS+, Kentrox, as well as other boxes for bridging
> protocol stacks.  We have not yet artificially insert 70 msec delay.  With
> your request, we will get on to that ASAP.  One question here:  since we
> are happy with what we found out so far in the lab it would be nice if
> we move the tests closer and closer to the real environment.  I heard you
> have a test network.  Can I connect my NAP prototype at Morristown NJ
> to your test network to study the effect of cross country delay?
>  
> Liang

Sure - you are buying the DS3 circuit - right.

Would you like to bring some of this equipment to NY?  We are not far
from the Tapan Zee Bridge, so you could easily drive some equipment
over.  That might be a wee bit more pracrtical than pulling a DS3.

Just for the record, Eric Bennet asked me about testing the NAP
configurations at least 3-4 months ago and I pointed out that egress
bottlenecks and cross country delays should be considered.  Even if
the traffic doesn't appear day one, you want to know if you have a
potential for collapse as load increases.

We have recently run performance tests using multiple TCP flows and
the equipment we currently use.  We currently have a 68 msec and 20
msec path, though we may lose the 68 msec path soon.  If you help set
up, I can do 2 automated 12 hour runs, one at 20 msec, one at 68 msec,
and we can do a direct comparison.

Doesn't Belcore have some form of delay simulator for DS3?  Such
things do exist.

BTW - If the 10 SSs are not running RFC1323 and using large windows
over a long delay you will see much less burstiness than in the tests
we have been running.

If you are interested we can take this discussion off line and work
out the logistics.  It would be in the interest of a smooth transition
to know what the limits are.  There may be cause for concern or no
problem at all.

Curtis