[563] in Pthreads mailing list archive
Why I need it for Solaris
daemon@ATHENA.MIT.EDU (Neil Radisch)
Thu Jan 30 07:13:19 1997
From: Neil Radisch<nradisch@duck.com>
To: pthreads@MIT.EDU
Date: Thu, 30 Jan 1997 06:39:00 -0500
Cc: nradisch@duck.com
I've gotten quite a few e-mails asking why I'm bothering
to compile Proven's pthreads for Solaris, when Solaris comes
with a pthread library.
Unfortunately the SUN pthread library is not complete.
In particular, you have no control over thread scheduling.
The program I'm building does realtime processing
and requires round robin scheduling so
that the processor time is spread evenly over the threads.
With the Sun threads you are at the mercy of their
scheduling algorithm which does not divide up
the time very evenly amongst highly active threads.
Even if the threads call pthread_yield(), you are
not guaranteed anything. The following quote
is from an e-mail sent to me by an engineer at SUN
>>thr_yield() is NOT a reliable way to get fairness.
>>It is seductive in its simplicity, and but doesn't
>>guarantee to run the thread you expect.
I translate that to "it doesn't work right".
In the long term we plan on redesigning our software so
it is less sensitive to thread scheduling issues, but we need
to get this thing working on a Sun NOW.
Thanks for listening.
Neil
------------
Neil Radisch
nradisch@duck.com
