1) Message boards : Number crunching : Suggestion on work unit distribution (Message 9679)
Posted 27 Aug 2005 by Mibe, ZX-81 16kb
Post:
Since the work in this project is generated in a different way compared to forinstance SETI-II, I think it is possible to have a somewhat different way of distributing wu's.

To my understanding LHC:
- sends out a batch of wu's
- wait for the result from all the wu's
- the scientists evaluate the results
- and then a new batch is generated based on the previous result and sent out.

If it's important for this project to have the work finished in a timely manner. I would suggest to change the way wu's are sent out.

The estimated time to complete all the wu's can be used to have a treshold after which the maximum number of wu's sent to each cruncher is decreased. Until at the very end only 1 wu is sent to each cruncher.

With this max cap the complete project is finished in shortest time possible when the last wu's for this round is sent out. In the best scenario within one day after the last wu is sent out.

An example: 100'000 wu's are generated and crunchers start download wu's with theirs individual setting on cache size etc. Some dl 1 wu others dl more. When the wu's are starting to dwindle and the treshold is reached the crunchers that usually dl a lot of wu's gets a little less. This is repeated until at the end when there are as many wu's left as there are crunchers every one only gets one wu. The end result is that when the last wu is dl'd there is only one wu in every cruncher and they should return within the slowest crunchers crunch-time, perhaps within one day.
2) Message boards : Number crunching : This guy really needs to check his computer, or overclock less... (Message 9191)
Posted 6 Aug 2005 by Mibe, ZX-81 16kb
Post:
I got a similar problem. The .exe can't start so all wu's are reported invalid.

There's another strange thing; when I try to dl a new .exe it takes a couple of days to complete the dl. During this time the CC tries to compute a wu between every dl attempt and fails of course. This failed attempt is reported and a new wu is fetched. Should'nt the CC wait until the .exe is completely dl'ed before starting the computation?



©2024 CERN