Hi Tracy,

I'ld first use the sledge hammer approach to verify bottleneck location:
install the app on the "underpowered" Celeron's local disk and remove it
from the network <g>. If they performance is adequate,
you know you can leave this hardware alone.

(background: the PIV-Celerons got a bigger boost from doubling the cache from
128 to 256 than the PIV from 512 on upward. The current celeron is much nearer
in performance to the PIV, although still measurably slower.)

If the Celeron is adequate with local install, run the app locally again
under coverage profiler. Run the app accessing the server under coverage profiler.
Search for the areas where the difference is greatest - don't make a full analysis.

use filemon as Walter suggests,
sniff the network
just look at the code <g>,

This is a perfect way for me to find suboptimal code which is not hurting enough on the desktop - I have a 10MBit switch just for creating such a (stable) bottleneck <bg>. It also is a much better approximation of the UC of the user's laptop with the 30GB disk filled to the brim with a lunatic partitioning schema on top and *never* defragged, so every hard disk access is similar to snail racing.

Before upgrading the whole network to 1GB, check if perhaps only the server card is flooded: then a GB connection from server to switch might be enough, since the switch can distribute to quite a few WS more.


HTH

thomas