Hi Jim,

it mostly basic stuff which wasn't done when the group
of 8 pc's was set up. These were quite niche machines
("back then")
PIII 800,
1GB of Ram,
1 5400 IDE,
1 7200 IDE
Nt 4 SP5

upgraded to current drivers for graphics, LAN,
and most significant: busmaster IDE,
which is NOT used by default in NT4.

Switched LAN from TCP/IP to NetBEUI
(was a physically closed net for security reasons already,
so catch the better performance in a small networks)
upgraded to SP6a, reset the size of filesystem cache
since the machines had lots of RAM not used by the
(FPW!!) application. Set NT not to swap out, since
most of the memory was still unused.
USED the 2 different drives to balance IO
and reduce head movement.

That cut exectuion times of the smallest datasets
from 84H to 48H. Since the execution times grew
nonlinearly with the data set size, I put together
a small profiling tool in 16-bit c++ and found
the relevant calls. I then optimized (mostly in the
searching routines) and cut execution times down
by a factor between 8 and (estimated) 15,
depending on the dataset size.

Since this was a legacy app converting host data
to FPW data structures and now used with very large
data samples for datamining purposes, this result
was "good enough" for quarterly operations.

Regards,

thomas