ejb@think.COM (Erik Bailey) (06/10/86)
Gasp. Choke. Even net stuff is too cruddy for me. The line-eater dies... Y'know what I think would be a neat (and useful) proram for the PC? The equivalent of a minicomputer CTL-T. I mean, I like to know what my computer is doing. CTL-T is great when the system load is high on the host, but there have also been times when I've wondered what my poor compaq deskpro is up to... Anyone want to write such a beast? How 'bout figgerin' out a neat estimate of load average? This could be cool! Just a thought... --Erik -- Erik Bailey -- 7 Oak Knoll (USENET courtesy of ihnp4!think!ejb Arlington, MA 02174 Thinking Machines Corp. ejb@godot.think.com (617) 643-0732 Cambridge, MA)
bill@hpcvlo (06/16/86)
*Why* would anyone want to write such a beast? I was a TOPS-20 user for years, but now do almost everything exclusively on MS-DOS machines. While I found ^T to be very useful on the timesharing/multitasking DEC-20, I don't really know what useful information I'd get out of it on a single- user DOS system. I suppose it could tell me that 99.9999% of the time it's sitting in an idle loop, waiting for keyhits... or that the CPU is either 0% busy or 100% busy. YOU are the only real "system load". And if you can't figure out what you're doing, what good is ^T going to do for you? bill frolik hp-pcd!bill
jrv@siemens.UUCP (06/20/86)
Maybe what the poster is asking for is an execution profiler. After hitting
the RETURN key and watching and waiting he might wonder what the poor
compaq is doing *while executing* his program. My use of Cntl-T mostly
occurs as a futile answer for my question "Why is this program taking so
long?" Cntl-T "Ah, the system is attempting to run well beyond its
capabilities. Where did I put that magazine?"
Tuning your program is something which few people seem to do. A few months
ago someone posted a request for techniques used for tuning programs.
I was the only response he got and I was asking for him to forward the
answers he got. For a while I was associated with a program widely known
as being a DOG. It was 200K of FORTRAN code and 350K of data. After a
little work with a logic analyzer to profile the execution of the program
we found that 25% of the time was spent in 5 lines of FORTRAN. By reworking
the algorithm used and putting in some judious assembly language the program
was speed up by 20%. Not an order of magnitude but still helpfull.
There are several software profilers available to run on IBM and compatible
computers. All of them take their toll on execution speed. A hardware
assisted approach when allow full speed execution but at much higher cost.
Jim Vallino
Siemens Research and Technology Lab.
Princeton, NJ
{allegra,ihnp4,seismo,philabs}!princeton!siemens!jrv