Re: Round trip query to the X server.

New Message Reply Date view Thread view Subject view Author view

Phil Keslin (philk++at++cthulhu.engr.sgi.com)
Thu, 18 Nov 1999 14:46:43 -0800

Bernard Leclerc wrote:
>
> Phil Keslin wrote:
>
> > At present, Performer does not track the viewport state. This requires
> > that the viewport be reset every frame (and hence the window geometry is
> > retrieved every frame). We have looked at fixing this in the past, but
> > haven't addressed it yet.
>
> Phil, it's been a week since I posted my original message to
> info-performer. Since then, a colleague suggested I look into promoting
> the X server priority to the real-time band. It works well.
>
> But I still think that GL and/or Performer should perform caching on the
> viewport setting and avoid doing unnecessary round trip request to the X
> server. Is there a plan to address this problem?

Not that I know of. I don't work directly on Performer any longer, but I'm sure
that Jenny and her team will take your suggestion under advisement.

> In the meantime, if anyone is interested, here are the steps required to
> ensure pretty damn good response from your O2.
>
> 1- Boost the X server priority
>
> $ npri -s RR -r 120 -p XsgiPid
>
> 2- Also boost the window manager priority
>
> $ npri -s RR -r 20 -p 4DwmPid
>
> 3- In a normal shell, run "osview -i1" to observe the swapbuffer rate.
>
> 4- Try the simplest Performer program such as simple.C
>
> $ npri -s RR -r 10 simple x29.sv

You must be very careful setting priorities in the realtime band. For the
Xserver, you have set the priority higher then several of the kernel tasks (at
least on IRIX 6.5). This could lead to deadlock (although I can't think of why,
its always possible). For more information on what gets placed were in the
realtime band, checkout the realtime(5) man page.

> Notice how stable the output is even under heavy system activity. For
> instance, on my machine, a low priority background task uses all the CPU
> spare time. So basically, my O2 has 0% spare CPU time. Even under this
> heavy load, the above steps allow a simple Performer program to run
> close to real-time. The example above runs at a steady 72 Hz on my O2.

- Phil

-- 
Phil Keslin <philk++at++engr.sgi.com>

New Message Reply Date view Thread view Subject view Author view

This archive was generated by hypermail 2.0b2 on Thu Nov 18 1999 - 14:46:55 PST

This message has been cleansed for anti-spam protection. Replace '++at++' in any mail addresses with the '@' symbol.