Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Seventeen or Bust :
Spliting SoB WUs in smaller pieces?
| Author |
Message |
|---|
|
|
|
Hi,
Is there a posibility to split the SoB WUs in smaller pieces/WUs?
Imho this would have some advantages:
- people which crunch only some hours per day/per week can contribute to SoB
- if a small WU is lost (error/hardware crash ...) the 'wasted' time is lower compared to a full WU
Maybe it is possible to use some checkboxes in the projectselection like:
Send full WU = Yes/No
Send small WU =Yes/No
____________
 | |
|
|
|
|
It would be hard to do, this means a whole rewrite of the software. You are doing a complete check of a single number and to split that means it would have to send a lot of data back to the server, the server would have to then create a new unit out of that data and then send that out. It would take a lot longer to get a completed unit. Plus the people that get the first part(s) would never find a prime as you are just doing part of the work.
Just not conceivable.
____________
My lucky numbers are 121*2^4553899-1 and 3756801695685*2^666669±1
My movie https://vimeo.com/manage/videos/502242 | |
|
|
|
|
I get your points,
my idea came from the fact that the SoB-Client (and mprime) are able to pause and resume the work on a candidate so I thought it would be 'easy' to stop the job after a certain amount of iterations and send the needed informations back to the server which can resend these informations as a new WU (I have absolutely no idea about the amount of work this would cause).
For those who want the chance to find a prime there could be the WU-switch I mentioned.
____________
| |
|
Honza Volunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,149,297,698
RAC: 2,302,314
                                      
|
|
This it not the first time such a suggestion was experessed.
See this post in "Send WUs to 3 users instead of 2?" for example.
As PB pointed, it would need rewritting a lot of aplication and BOINC server code, quite complex.
And potentional gain is questionable.
Another approach is to make LLR running on GPU. This has been tested but benefit was not there - it wasn't consdiderable fast and still used CPU code to feed GPU.
Yet another approach is to make LLR application (that does SoB test) multi-threaded.
A single test would run shorted time - 1/2, 1/4 or even less time depending on number of cores and overall effeciency of application.
But this would also need a considerable rewrite of LLR app, extensive testing and some adjustemnt to the BOINC server side.
Above all, PrimeGrid project is not author of LLR but rather a contributor and intensive user.
____________
My stats | |
|
|
|
|
OK, thank you both for the answers.
I thought that it is 'only' necesarry to change the behavior from the application.
But it looks like there is a lot more to do. thx.
____________
| |
|
 Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 435,948,523
RAC: 876,383
                           
|
OK, thank you both for the answers.
I thought that it is 'only' necesarry to change the behavior from the application.
But it looks like there is a lot more to do. thx.
Unlike some other projects, you can't just make these tasks longer or shorter. We can with sieves, but not with primality tests. It's already as short as possible: it's doing just one test.
It's theoretically possible to split the single test into multiple parts, but at tremendous cost in terms of disk storage and bandwidth, since we'd have to move very large checkpoint files back and forth and store them on the server. Additionally, the LLR program (which is written by an outside party) would need to be modified to stop after a specific number of iterations.
So while it could in theory be done, it's extremely unlikely to happen.
____________
My lucky number is 75898524288+1 | |
|
|
|
|
CUDA 6.5 is supposed to have some enhancements that a GPU LLR may benefit from. I know it's not a top priority, but something the developers could look at when time permits.
| |
|
Message boards :
Seventeen or Bust :
Spliting SoB WUs in smaller pieces? |
