Discussion:
run-program and performance
Paul Griffioen
2014-08-04 00:10:15 UTC
Permalink
Hi everybody,

I have some performance issues when a program is run with ccl:run-program
on Windows, and it looks like the communication to the program is slow. Is
there anyway I can check whether buffering on the streams is on or off?

I am guessing at this moment, but I just want to make sure I am not
overlooking something obvious. See
https://gist.github.com/anonymous/80966ad2b1011de4a75e for the code.

It runs fine on Linux. Similar code also runs fine in other Lisps on
Windows. Any clues would be appreciated.

Paul
Paul Griffioen
2014-08-06 10:32:57 UTC
Permalink
Thanks for the reply.

In my case the communication is mainly from lisp to the program.

As I wrote, I am not really certain what is causing this and I am trying to
eliminate obvious candidates. Now at least I know that the streams are
buffered.

It seems not so easy to pinpoint this issue. I am running ccl from
binaries. Maybe I can run ccl from source and see if I can do some
experiments, or do some profiling. Would you have any suggestions? Or any
specific relevant ccl code I could look at?
Is communication slow from the program to lisp, from lisp to the program,
or in both directions ?
Note that (almost) all streams in CCL are fully buffered, i.e., output
may not actually be written to an underlyin device (file descriptor, file
handle)
until FORCE-OUTPUT is called; this is the same on Windows as on other
platforms,
Post by Paul Griffioen
Hi everybody,
I have some performance issues when a program is run with ccl:run-program
on Windows, and it looks like the communication to the program is slow. Is
there anyway I can check whether buffering on the streams is on or off?
I am guessing at this moment, but I just want to make sure I am not
overlooking something obvious. See
https://gist.github.com/anonymous/80966ad2b1011de4a75e for the code.
It runs fine on Linux. Similar code also runs fine in other Lisps on
Windows. Any clues would be appreciated.
Paul
------quoted attachment------
Post by Paul Griffioen
_______________________________________________
Openmcl-devel mailing list
http://lists.clozure.com/mailman/listinfo/openmcl-devel
Loading...