Russell Coker wrote in a message to Henrik Falk:

 RC>    I believe that OS/2 doesn't guarantee fair queueing for
 RC> threads (although I expect that it makes at least a minimal
 RC> attempt at reasonably fair queueing).  This means that if
 RC> you have 2 players in a game with seperate threads then on
 RC> occasion one will get an advantage over the other due to
 RC> OS/2's decisions on which thread gets priority.  This means
 RC> that the game will not always play fair if it relies on OS/2
 RC> to give equal timeslices to each thread.

This is exactly right.  In fact, the effect will be extremely unfair if the
processes are usually both waiting on I/O, since the I/O will usually not
be assigned in round robin fashion.  This can be compensated for if you
have control of every thread on the scheduler by making explicit yield
calls, but that will also hurt performance, achieving fairness at the
expense of thrash.

 RC>    I find it difficult to conceive of a good reason for
 RC> having more than 1 thread for calculations related to moving
 RC> characters etc in a game. 

It depends on the application.  One thread per I/O device is also a
reasonable way of divvying up the work, and I have often created five or
six threads to cope with a similar number of I/O devices.  I still like the
rather subtle briliance of Peter Fitzsimmons' rule that "the job of a
thread is to do nothing very well."

 RC>  X MR/2 2.0 NR X Ever wonder what Wonder Woman wonders about? 

Superman?
 
-- Mike


---