Short work units
24 Aug 2006 3:36:21 UTC
Topic 191730
(moderation:
Ive read somewhere that the slowest 10% of hosts get smaller work units. However, my P3 seems to keep getting long work units.
Why is it that some hosts that have higher benchmarks are getting small wus consistently?
Thx
Short work units
)
Basically, it's luck of the draw. The server tries to skew the probabilities in favor of assigning a small wu master data file to slower hosts, but it isn't perfect. So, if you received a long wu master data file, you will likely have to wait until all 'slices' of that data file have been crunched before you will request a new one. At that time, you will probably receive one with small wu slices if you are still in the bottom 10% of hosts.
-- Tony
Yep, my PIII 700 currently is
)
Yep, my PIII 700 currently is doing a batch of large workunits, taking roughly 21.5 hours each. Still gets the credits though :)
RE: Ive read somewhere that
)
See also this thread, especially message http://einsteinathome.org/node/191541&nowrap=true#46010
and this thread where Bruce describes the algorithm with which Boinc selects long/short running WUs.
Udo
Udo
I just noticed this
)
I just noticed this thread.
I have a K6-2/500 which is running a large data pack at the moment and it takes around 9 days to finish a result running 24/7, but it drew this when S5 started. Slow but steady. :-)
Based on the post from Dr. Allen, it would would seem now you need to have performance roughly equivalent to a PIII 1.0 GHz to be able to draw a long data pack.
The thought just occured to me that you need to take into account your overall BOINC project load as well. IOW, an even faster machine running a lot of projects may be precluded from drawing long packs depending on what its resource share happens to be.
On second thought, the problem with a heavy project load machine would not be that it wouldn't draw a long data pack, but wouldn't be able to finish a result in time without having to go into EDF mode, periodically.
Alinator