Message boards : Number crunching : Validation
Message board moderation

To post messages, you must log in.

AuthorMessage
JIM

Send message
Joined: 22 Feb 14
Posts: 4
Credit: 24,911
RAC: 0
Message 26215 - Posted: 25 Feb 2014, 16:25:46 UTC

I have just started running LHC@home and I have a concern. It deals with the validation. While reading about this project online I came across a posting (about 3 years old), that if I understand it right, said that 5 copies of each WU are sent out, but, only the first to finish 3 are needed to validate. It said that the last 2 results are discarded as unnecessary. Is this true?

This is a concern to me as I don’t have the fastest machines and I don’t want to waste my time. Are the 4th or 5th place finishes just tossed into the trash can as unneeded.

The posting that I read also said that the moderators remove questions like this because the project Admins don’t want to admit that this wasteful practice exists. I am used to running climate models over at CPDN. They except everything that is finished, even if it arrives 3 years after the deadline.
ID: 26215 · Report as offensive     Reply Quote
henry

Send message
Joined: 15 Sep 13
Posts: 73
Credit: 5,763
RAC: 0
Message 26216 - Posted: 25 Feb 2014, 17:11:36 UTC - in response to Message 26215.  

They've stopped that very wasteful practice. They issue 2 copies of each task and don't issue a 3rd copy unless the first 2 results don't match. They need 2 matching results to verify that they have the correct "answer" for the calculations in the task.



6 bangers are for wussies.
ID: 26216 · Report as offensive     Reply Quote
JIM

Send message
Joined: 22 Feb 14
Posts: 4
Credit: 24,911
RAC: 0
Message 26217 - Posted: 26 Feb 2014, 14:49:44 UTC

Thank you for the info. I am very happy to learn that they no longer use the wasteful 5/3 system. I can now have confidence that my efforts will be worth the electricity.

ID: 26217 · Report as offensive     Reply Quote
Eric Mcintosh
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 12 Jul 11
Posts: 857
Credit: 1,619,050
RAC: 0
Message 26218 - Posted: 26 Feb 2014, 15:48:14 UTC - in response to Message 26215.  

Don't worry Jim; we are not 'cheating" like that.
We run only two copies, the minimum, and only run additional
WU's when there is a discrepancy.

It is true that we did run many more a long time in the past
when starting up and actually forgot! to reduce the number until
a friendly volunteer pointed it out.

Thanks for your interest and support. Eric.
ID: 26218 · Report as offensive     Reply Quote

Message boards : Number crunching : Validation


©2024 CERN