TPS benchmarks for an active system

Can anyone suggest a way to access near-real-time transaction per second
benchmark information from an active production installation of Postgres
7.1?

We have a 100+ user OLTP application and are about to implement some
hardware changes; moving log file to its own spindle, raid 1, etc.  What
I'm hoping for is a method by which to look into the functioning production
system under load and have a way of saying that on average, it's currently
processing X tps.  This would provide a benchmark for comparison for after
the hardware upgrades.  It would also provide a means of gauge increased
use/load over time as more users and/or increased functionality is
deployed.  This in turn could help us plan for (and justify to management)
the need to boost platform hardware IN ADVANCE of reported performance
problems.

Basically, I'd like to be able to know how many transactions per second
were processed between 8:45 and 9:00am, between 1:30 and 1:45pm and between
4:00 and 4:15pm and track these number historically for comparison
proposes.

I've done this in other environments where data from the logging system of
transaction starts and transaction ends could be captured, accumulated and
used to come up with these numbers.  With the sophistication of the
Postgres WAL subsystem, I'm sure the information exists.  It's just a
matter of knowing the questions to ask in retrieving/accumulating the
required data.

Anyone have any ideas?

Marc Mitchell - Senior Application Architect
Enterprise Information Solutions, Inc.
Downers Grove, IL 60515
marcm@eisolution.com