[RC5] OGR and CPU time
oetting at gldmutt.cr.usgs.gov
Wed Feb 23 14:10:11 EST 2000
At 6:04 -0700 2/23/2000, Jonas Maebe wrote:
>>I run OGR on a G3 Macintosh. Ever since I started OGR I have noticed that
>>having the client running in the background noticeably slows up other
>>programs I am running. RC5 and CSC never had this kind of effect on my
>>computer. The result is that I am pretty much unable to run OGR while
>>actually working at my computer. Has anybody else experienced a problem of
>>this type, or is it (somehow) just me?
>No, I experience it too. What's strange is that it isn't that way
>continuously, but only most of the time.
In RC5 we know exactly how many machine instructions are required to
process each key so we can easily compute the number of keys to process
before we need to check for user events again. In OGR the work that gets
performed for 1 iteration can vary significantly. In addition, the OGR core
doesn't even contemplate yielding while it is processing marks in the
second half of a ruler. This results in the erratic and unacceptable
responsiveness you are reporting.
The good news is that this problem can be fixed. By using a time event
based yielding we can more closely mimic preemptive scheduling. When it's
done properly you won't even know the client is running in the background.
The Mac coders are working on this problem. In the mean time, if you want
to see how time event based yielding works you could run Rob Harley's
ecliptic curve client <http://cristal.inria.fr/~harley/ecdl7/>. Certicom is
putting up the usual prize and I estimate that the solution will fall in
about 2 weeks now that the rest of the group has done most of the work. And
check out the speed of the new G4 client on the ECDL speed page
-- Dan Oetting <oetting at ghtmail.cr.usgs.gov>
To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
rc5-digest subscribers replace rc5 with rc5-digest
More information about the rc5