Re this:---
----- Original Message -----
From: "Peter Smithson" <[log in to unmask]>
> UNIX seems to be setup to read a byte at a time and MPE likes things in
> big chunks.  I've had problems on MPE getting it to read one char at a
> time so I think it's just that block mode is more appropriate for MPE and
> not for UNIX.

Having seen the improvements that come from MR-NOBUF reads on MPE (10 fold
or greater throughput IIRC), I wrote some C code on UX once, getting cpu
time consumed reading 1 mb of disk data in several different ways, each one
of them assuming the user wanted access to 100 byte logical records:-
100byte reads
1000byte reads, chopping up into 100s
10k  .....
20k...
30k...
40k..
50k..
etc
100k.....
1mb....chopping up into 100s.

As you would expect, the CPU time went down, but there seemed to be a knee
in the curve, best results being between 50 and 70k, IIRC. Maybe related to
system load, ram size etc, this was a few years ago. I got very similar
results under windows as well.
Anyway, the upshot is that UX will also benefit from the MR-NOBUF concept
for doing large serial reads of data, may just need some eveluation case by
case.

jp

* To join/leave the list, search archives, change list settings, *
* etc., please visit http://raven.utc.edu/archives/hp3000-l.html *