BOINC: A summary

[TAandy]

Janus over on the AP forum was kind enough to provide me with this summary of BOINC/Astropulse.
Hope it answers a few questions

">what we're doing at the moment

David and his team of developers are developing a framework for distributed computing. This means that in the future other scientists won't have to start over building everything up from the very bottom - they can use BOINC.

Currently we are trying to find any bugs in the software and improve the credit system so that "the game" is fair so to speak. Everything has to be checked and doublechecked before it is made available to all those thousands of thousands of potential users. We sure don't want to start with a crash =)

Also, we are waiting for the first port of SETI to BOINC to be finished (1-2 months from now). SETI will very possibly be the first project for BOINC (as Astropulse is delayed - or at least that is what I have heard...).

>and what we will be doing in the future

SETI will have a short phase where it is running on both the old client and the new BOINC version. After a while Astropulse will join, then SETI-II southern hemisphere project.

>an explanation of how the credits are worked out

The creditsystem is very different from the old SETI system. It prevents cheaters from having any personal gain by cheating (no credit) - so finally the competition is completely fair!
In the old SETI credit was measured in uploaded WUs - but sometimes a WU takes longer than others... Now in BOINC credit is measured in seconds of processing on a reference computer. This means that even though different projects have different WU times, the credit can be compared.

It works this way:

- A WU is given to 3 or more clients.
Let's say that you are one of these. You recieve a WU, process it and upload it again.
- On the homepage you will be able to see that your computer claims a certain amount of Cobblestones* (credit) for processing the WU.
- Now the 2 other clients return their results.
- The results are compared to check for cheaters/errors.
- If 2 or more (the majority) of the results are identical these will get credit (because they are considered valid). Wrong results do not get credit.

That is the way it should work. If something goes wrong in one of the clients, the WU is simply distributed to more users in order to validate the result.


If you have any questions, please ask =)


*Cobblestone: 1 day of work on the reference machine."

Hope this helps
Edit:
The reason people don't seem to be getting credit at the moment is because of the small userbase. The wu's aren't getting passed around quickly enough.

 

[lane42]

Thanks Andy, Thats a good read.

Question, Will sneakernetting work the same way. Most of my computers
have no internet connection so i load them on zip to upload.

 

[TAandy]

Originally posted by: lane42
Thanks Andy, Thats a good read.

Question, Will sneakernetting work the same way. Most of my computers
have no internet connection so i load them on zip to upload.

I don't think it will, at least not for a while
Have a look
here

 

[JWMiddleton]

Interesting read! Thanks TAandy!

Part of the discussion is about being able to move WUs on CD-Rs, zip disks, etc. Isn't that how we got into trouble with cheating in the first place? It looks as if a machine must have an Internet connection and download their own WUs. I know that will be a problem for some, but I don't see them being able to change that requirement.

 

[lane42]

This is great..............I spend thousands to help the seti cause and there screwing me :|

Anyother project need 16 Gigs of processing power, let me know.

 

[Coolkid]

Thanks 4 the info TAandy
Will try it again when i get home...

 

[Rattledagger]

Since you didn't mention it, one other important stats-fact:

Of the 2 or 3 validated results, all get the LOWEST credit.
AFAIK, if someone later returns the same wu and passes the validation, he gets the same credit even he maybe has lower "claimed credit".


Compared to Seti's real-time-stats, BOINC will be slow. I'm not sure how many days it's possible to cache, but I guess if you'll "lucky" all wu you crunch is also given to the 2 with max cache... so don't except credit the 1st month.
In the beta with that, 700 users, of course it's slow.

 

[Swanny]

So with the credit given by seconds of CPU time is there any advantage to having a faster CPU? If all that counts in seconds of CPU time, it seems to me that my computer with an Athlon XP 2500+ would generate the same number of "points" as my computer with a PIII 500Mhz. Is this true? If so, how is this considered "fair"?

 

[Rattledagger]

"claimed credit" = "cpu-time" * "machine-score"

Machine-score is based on the benchmark that's done of integer, floating-point & memory-bandwith 1st time you're starting BOINC, and atleast for the beta you can re-run this as often as you want.
I've not got a clue how "machine-score" is calculated based on these 3 numbers, or if more of the computer-internals is used to get this.

So basically, if you've got a 486-33 MHz that's using 1000 minutes to crunch a wu or a 486-33 GHz that's using 1 minute, you should get the same "claimed credit".

Since this is still beta, the machine-score will probably be changed yet again... I remember 1st time I fired up BOINC, and got negative floating-point-numbers...



UD/Think uses something similar, except that here it's run-time not cpu-time:
credit = "run-time" * "machine-score" / 5

 

[TAandy]

Recently posted on the AP forum

"Folding@Home (another distributed computing project that wants to use BOINC) gave us some insight towards credit systems.

We will be changing the range Cobblestone credit claimed/granted so that instead of getting say 0.15 Cobblestones you receive perhaps on the order of 100 credits. This will involve changing how fast our "standard" computer is. Outside of devaluing results before this change, it will have no effect except psychological."