In our benchmark of MaxConnect we found that the benefits began
showing up long before we hit the 150 user mark. However, as others
have pointed out, your environment may be different.

For our benchmark we used one of our standard database servers (AIX, 8
processor), a VERY small AIX server for the MaxConnect instance, and a
VERY small server to drive the user simulation scripts. We saw
reductions in overhead at the database server level and the
“users” saw a big reduction in total turnaround time on
the query (from SQL start to last record returned).

Our 10mbs network saturated but the database server hadn’t
broken a sweat. I was able to demonstrate to our network manager that
hops DO matter. We had to move around things to put together a test
environment so that no point was move the 2 hops from the next point.

One thing that wasn’t clear from your description: Was the user
simulation the equivalent of a Connect, fire one SQL, and a
Disconnect? If so, I can see why things got worse. While we
didn’t specifically benchmark this scenario, it would only make
sense that connection times would be slightly longer – though
that wouldn’t explain the 15 seconds to 100 seconds difference.

Fred Prose
fprose at supreme dot sp dot state dot az dot us
Arizona Supreme Court