*BSD News Article 28458


Return to BSD News archive

Xref: sserve comp.os.386bsd.apps:1037 comp.os.linux.misc:10964
Newsgroups: comp.os.386bsd.apps,comp.os.linux.misc
Path: sserve!newshost.anu.edu.au!munnari.oz.au!constellation!qns1.qns.com!news.sprintlink.net!demon!uknet!pipex!howland.reston.ans.net!wupost!csus.edu!netcom.com!hasty
From: hasty@netcom.com (Amancio Hasty Jr)
Subject: Re: DOOM for X
Message-ID: <hastyCMGpA7.Gu5@netcom.com>
Organization: Netcom Online Communications Services (408-241-9760 login: guest)
References: <2lkpg5$8hn@Tut.MsState.Edu> <hastyCMEw58.FMr@netcom.com> <1994Mar10.123047.15912@swan.pyr>
Date: Thu, 10 Mar 1994 18:42:07 GMT
Lines: 53

In article <1994Mar10.123047.15912@swan.pyr> iiitac@swan.pyr (Alan Cox) writes:
>In article <hastyCMEw58.FMr@netcom.com> hasty@netcom.com (Amancio Hasty Jr) writes:
>>I really doubt that a direct access to a fast svga card on a local bus would
>>compare in performance to the client-server model of X.
>I thought the same, but time it one day. XShm is incredibly quick, you
>just tell the x server do this shm object and carry on. On something like a 
>6 cpu sparc its even better as X is rendering your image while you work
>on the next one
>>
>>If you ask me X is very, very broken in this respect. 
>Apparently not.
>
Apparently yes, any client-server model which forces you to always have
two processes to communicate with each other is a broken architecture.


BTW: I was the one who went out and got all the little pieces to implement
Xshm for 386bsd....

Jon Tombs and I improved image read/write to the S3 cards by a factor 
of 10 by improving the routines which did the bulk transfer to the 
card. 

It is significant faster to download an image to a vga than to load
the image to a shared memory segment, make an IPC call, do a process
context switch, go through X and then finally download the image
to the card. 

Also, it depends on the application with mpeg_play and my P66 there
was no performance improvement  using  Xsmh, X was the bottleneck.
I managed to get 27 frames per second with no dithering and I got
17 frames per second  with Xshm. So it was either X or my S3 928
card. The S3 928 card is very comparable in performance to 
sun's gx.

My guess is that Xshm performance is directly proportional to the
size of the buffer upto the overall bandwith of the svga card.
Some of these new sgva cards internally have in excess of 240 mega byte/sec
transfer rate so their bandwith is limited by the bus interface.

	Amancio







-- 
FREE unix, gcc, tcp/ip, X, open-look, interviews, tcl/tk, MIME, midi, sound
at  freebsd.cdrom.com:/pub/FreeBSD
Amancio Hasty,  Consultant |
Home: (415) 495-3046       |  
e-mail hasty@netcom.com	   |  ftp-site depository of all my work:    
ahasty@cisco.com           |  sunvis.rtpnc.epa.gov:/pub/386bsd/X