home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!haven.umd.edu!darwin.sura.net!spool.mu.edu!olivea!sgigate!odin!twilight!zuni!anchor!olson
- From: olson@anchor.esd.sgi.com (Dave Olson)
- Newsgroups: comp.sys.sgi
- Subject: Re: Real time features/problems under Irix 4.0.4
- Message-ID: <peocgpo@zuni.esd.sgi.com>
- Date: 5 Sep 92 00:17:36 GMT
- References: <1992Sep4.212119.13867@thunder.mcrcim.mcgill.edu>
- Sender: news@zuni.esd.sgi.com (Net News)
- Organization: Silicon Graphics, Inc. Mountain View, CA
- Lines: 44
-
- In <1992Sep4.212119.13867@thunder.mcrcim.mcgill.edu> root@amerie.mcrcim.mcgill.edu (Thunder Root) writes:
- | We are using a Personal Iris 4D/35, running IRIX 4.0.4, to run a
- | real time control task at 50 Hz.
- |
- | The real time task is created using sproc(), and given a non-degrading
- | process priority using schedctl(NDPRI, 0, NDPHIMAX). All memory in the
- | program is locked using plock(PROCLOCK), and stack space is locked as
- | well using mpin(). The real time event is generated by a VME bus
- | interrupt that is caught by our own driver, which then calls the
- | kernel routine pollwakeup() to wake up the real-time control task
- | which is waiting inside a select().
- |
- | We can verify the performance of the task using an oscilloscope.
- | It runs fairly reliably (with a maximum jitter of 2 or 3 msec.) if I
- | edit files, run make, or list directories. However, serious delays
- | of around 200 msec. or so can be caused by things involving the
- | screen display -- such as moving and exposing windows. Scrolling
- | inside an xterm window will also cause such a delay (although
- | scrolling inside a winterm does not).
- |
- | Our documentation on IRIX real-time talks about guaranteed process
- | dispatch times of 1 to 2 milliseconds for real time tasks. So ...
-
- That is on the MP machines (with more than once CPU).
-
- | 1. What really _is_ the worst case process dispatch time?
-
- It hasn't been measured on the ESD platforms, because we have
- no official realtime support there (although many of the realtime
- features exist and are implemented).
-
- | 2. Is there anything we can do to improve ours? (short
- | of booting the machine single user :-))
-
- Don't run graphics ;)
-
- Seriously though, there are cases in the graphics code where it
- spins at interrupt time waiting for the graphics fifo to empty.
- There has been talk about fixing this for several years; it is
- a somewhat difficult problem to solve...
- --
- Let no one tell me that silence gives consent, | Dave Olson
- because whoever is silent dissents. | Silicon Graphics, Inc.
- Maria Isabel Barreno | olson@sgi.com
-