Re: Splitting pfPipes across several machines

New Message Reply Date view Thread view Subject view Author view

Jim Helman (jimh++at++surreal)
Tue, 21 Jun 94 11:30:13 -0700

   
> We have a customer who is interested in using Performer
> where each pfPipe is on a separate machine rather than
> being separate pipes in a single machine. Has anyone
> tried to do this?
   
Several things are required:

1) The hardware pipes should be genlocked.

2) You need a mechanism to synchronize the notions of time
(pfInitClock and pfInitVClock). This requires some network
mechanism for time synchronization, preferably down to a
around a millisecond. I think network time protocols like NTP
might come close. Once you have time of day and pfGetTime
synced to well under a video frame period, syncing the VClocks
can be done with a little communication over standard ethernet.

3) If you want visual coherency down to individual video
fields, you also need to synchronize swapbuffers so that the
displays won't get out of sync by a video field if one pipe
frame extends. You can do this pretty well using a very
low-latency interface such as SCRAMNet or a VMIC shared
memory, but there will still be a small "hairy edge" where
sync failure is possible. Reality Engine has a GANGSWAP wire
which allows synchronization of swapbuffers across multiple
pipes. The pfChanShare attribute PFCHAN_SWAPBUFFERS_HW
enables this. HOWEVER, this mode should only be used in
dedicated full screen applications. GANGSWAP cannot coexist
with cursor changes (e.g. crossing X window borders).

rgds,

-jim helman

jimh++at++surreal.asd.sgi.com
415/390-1151

New Message Reply Date view Thread view Subject view Author view

This archive was generated by hypermail 2.0b2 on Mon Aug 10 1998 - 17:50:21 PDT

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