[Prev][Next][Index][Thread]
Re: Executor on Xaccel & XFree
>>>>> "Jeff" == Jeffrey A Halverson <jeff@Ika.umn.edu> writes:
[interesting stats deleted]
Jeff> What I thought was particularly interesting was that
Jeff> calculations seem to actually speed up with AccelX also.
Jeff> Why is this? I was expecting those to stay constant, but
Jeff> only have the video-related tests to improve... Can anyone
Jeff> explain this odd phenomena?
Of the three primary Executor engineers, I'm the least qualified to
answer, but since nobody else is volunteering any info, here are my
thoughts.
I'm a little surprised, too. However, the way the program works is it
does a bunch of tests and sees how much time elapses during the test.
*During* the test speedometer does a little bit of graphics in that it
changes the cursor. In addition, there are some fancy graphics done
right before and right after the test and buffering may cause these to
also affect the times. Furthermore, depending on how much memory you
have, there could be paging that could affect things.
I don't know how much memory you have, but I'd guess paging and if not
paging, context switching overhead. It could be that XFree requires
many more context switches even when idle than AccelX.
Cotton might know.
Jeff> Jeff
--Cliff
ctm@ardi.com
References: