Security question?

Keith Packard keithp at keithp.com
Thu Nov 4 08:35:48 PST 2004


Around 10 o'clock on Nov 4, Alan Cox wrote:

> We have epoll() but I've seen no evidence it makes any difference at
> such a low connection count.

I've got two data points which suggest that we could see modest 
performance gains with the elimination of select:

 1)	The multi-threaded X server was faster than the single-threaded X 
	server when large numbers of clients were connected (it blocked in 
	read(2) and never used select(2)).

 2)	The current X server scheduling algorithm has a hack which extends
	the current client time incrementally from 20 to 200 ms to improve
	benchmark numbers. (yay benchmarks).  That's the interval between
	select(1) calls which serve to check for other client activity.

I'd like to make it possible for higher priority clients which become 
ready to interrupt the current client in the middle of the time slice.

I'd also like to reduce the slice so that multiple simultaneous animations
are smoother on the screen even when they are overrunning the available CPU
time.  That would require faster polling code and a better way of getting 
time of day into user mode.

-keith


-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
URL: <http://lists.x.org/archives/xorg/attachments/20041104/a4fe72bd/attachment.pgp>


More information about the xorg mailing list