As we watch the impact of the tropical storms hitting the US this year, I wonder how a projct like boinc could help.

We all know that the accuracy of any forcast is based on the level of detail that is processed. As the number of processing points increases, the cpu required shoots up, but so does the accuracy of any forcast.

So, when there is an urgent need for more power, why couldn't users allow their choice of projects to be paused while their machines worked on a current real time problem?

I know that you would need a way to ensure that the results were real, but that problem could be solved.

Ed

Makes sense... but the biggest variable is when will the data be returned? 1) How many duplicate work untis do you send out hoping to get a timely response on each and 2) will the data be useless by the time any meaningful response is received because of changing conditions?

BOINC is the world's largest, but uncontrolled, network/supercomputer.

I think those would be the biggest obstacles.

On the other hand, CPDN tackles weather in a (sort of) time insensitive way.

> On the other hand, CPDN tackles weather in a (sort of) time insensitive way.

Yea, i liked that, does take a while to compleate :)
I was in hurricane Charlie while i was on holiday :D i thought it was "kool"


^ Click Here ^ ^ Click Here ^ ^ Click Here ^

> Makes sense... but the biggest variable is when will the data be returned? 1)
> How many duplicate work untis do you send out hoping to get a timely response
> on each and 2) will the data be useless by the time any meaningful response is
> received because of changing conditions?
>
> BOINC is the world's largest, but uncontrolled, network/supercomputer.
>
> I think those would be the biggest obstacles.
>
> On the other hand, CPDN tackles weather in a (sort of) time insensitive way.
>
Personally I think ALL those issues could be remedied by a little forethought on who can participate in the project.
A user MUST be able to return the data within 6 hours of receiving it!
If #1 can be done then any unit not returned within that 6 hour time limit will be discarded, if your computer can't do this then no need to even apply. That means that the units must be small enough and the computers on the sending/receiving end must be able to handle the data. All units would be sent out say 10 times initially and then again if at least 4 don't come back in time. As the units are returned the data can be used with the caveat that they are not verified until more of the same pieces are returned.
My computer currently does Boinc units in about 2 hours, there are ALOT of users that can also do this time frame or less. That means that a splitter could be created and the data sent out and the first returns would start coming in in about 2 hours or less. That should give some initial reasonable data to be able to start the process of putting it all back together for the final analysis.

The initial problem would be how and where to split the data off into the chunks that we users crunch. A Scientist would have to have the data in hand to be able to decide that, but I believe Boinc would be an excellent place to start.
The second problem would be to send out new units based on the results of the inital set of work units. That would involve resplitting the units in new places so the model evolves in the indicated way and ignoring the returns that are still based on the old model.