peter @
nmti .
com wrote:
>I'd like to know what's wrong with the fwtk proxy as well. I'm using the
>plug-gw to an outside cern proxy.
One of the problems that we experienced with fwtk http-gw was that it had to
read its config file every time someone wanted to get any file download. So,
if you have a page with 4 pictures in it, each of those pictures would cause
a re-read of netperm-table. So each page with 4 inline images cost 5 reads
of the config file (1 for the original page + 4 for each inline image).
If I were to make a WAG (*), I'd say there are about 4 in line images per web
page. If 10 people are trying to download web pages at the same time, that's
50 copies of http-gw running, each reading netperm-table. Now, we have a
couple hundred people who access the web to get work related information on a
constant basis. During work hours, the load on the gateway averaged about 75
and peaked at about 150!!
(*) WAG = "Wild Ass Guess", blatently stolen from mjr.
We also saw that people who had their IP addresses near the top of
netperm-table would get significantly faster response time than those people
who had their names near the bottom. The behavior was testable; put someone
from the bottom of the list at the top of the list and their response time
became what you would expect. Put that person back on the bottom and the
response time fell through the floor.
That's the reason that I can't use http-gw. I like it fine, but its
archetecture just doesn't allow my users to get their work done.
--
Mark Horn
mhorn @
funb .
com
Free Advice and Opinions -- Refunds Available
References:
|
|
