[Knightrider]

There has been some debate elsewhere, that running an optimised manager or client is cheating in some way, but this is not my view.

BOINC is open source and issued on a lesser gnu licence, which as I understand it, allows the modification of these programmes. That said it is I presume, possible for Berkely to write into thier program the ability to determine which version of the manager and/or client being used and to inhibit it in some way.

The fact that the normal BOINC benchmarks are set up to provide a level playing field means that the faster you process wu's the lower the credit you are going to get, the slower you do them the more credit you get. So inherently there is no reward for increased effort. In my view this inhibits competition.

This was set up by Berkeley because of cheating going in SETI Classic and whilst it is a solution, I believe it is not the correct one for the reasons I have just given. So what would be a better solution?

IMHO it would be:-

1/ To raise the level of credits awarded to levels similar to those given by optimised versions.

and

2/ To adjust the levels of credit so that the faster a wu was completed the greater the reward.

Comments?


K.

[Todd Madson]

I never understood how people were cheating under Seti Classic. Can someone explain this?

I ran the stock client, slow though it was on a variety of machines and welcomed
the fact that faster processors or optimized clients were available.

The only way I could see cheating was involved was if someone had altered it so
that a work unit somehow was "worth" more than the value claimed.

I don't see it as a huge issue. I do see griping from people with older, slower
computers wanting just as much credit as those who have faster newer ones.

The idea of a level playing field is an honorable one but not achievable. There
are some out here who have thousands of members with thousands of computers.

You can't hope for a home based "farm team" of a handful of people running
five year old computers to realistically achieve the same results as a team of
2000 people with brand new computers.

At the same time, I understand how they feel but in real life things aren't always
fair.

I realize there are people out there who can blow my doors off without even
trying but ultimately it's not a goal to "be #1", it's a goal to achieve a science
goal with the appearance of having a race of some sort.

[beadman]

Well said, Todd.

I also agree with Knightrider's comments. As far as I'm concerned, over-clocking a computer is no different that the much more expensive method of buying a faster computer. Heck, If I had a few spare Millions of dollars I didn't know what to do with, I might buy a Cray to run SETI on, although I think it would be cheaper and a whole lot easier to just buy 10,000 G5 duo servers, like Virginia Tech did... I'd certainly like more speed, but I'm using the G4 Powerbook work gives me, my personal G4 iBook, and my old Blueberry iMac because they're the computers I have. I need to spend my available money on other things than a whole room full of computers. Don't get me wrong, I have no problem with other's doing that.

For me, it's a chance to do something scientific with my computers when they're not being used otherwise. With an optimized client, I can do more science in the same amount of time, so using an optimized client makes a lot of sense. I have to admit, though, I get a kick out of watching my total credits creep upward in the listings, but it's not a big loss to my self-esteem if my team standing begins to fall.

If someone gives you permission to run BOINC on their computer, well and good - that is neither illegal nor unethical, in my opinion. Putting the software on their computer without their knowledge, on the other hand, is definitely unethical and may be illegal (you're stealing their electricity, at a minimum...).

Changing anything in the Work Units to give oneself higher credit I also consider unethical - after all, we're trying to do science here, and modifying the WUs will potentially give us wrong answers.

To me, the bottom line is: if someone else has more computers than me, that's ok - "being fair" doesn't enter into it at all.

beadman

[lepetitmartien]

I really don't see the point. Someone crunches more, he asks for less points, though more often. Someone crunch less because of his computer, he asks for more points less often and you can't do a thing about it. In the end they only get the medium credit claimed in the 3 first coming in it's a way to average the credits to something statistically realistic over the current computer pool. You can't fake, you can only do more or less work for science, and it's not time which is given credit but some approximation on the job done. As long a the clients do good computing (actually the best they can, it's science that counts, not speed) they get credited on the job they do. If people were credited on the time they take, the interest would be to run at the lowest speed possible…

Not exactly what's needed, is it?

[reader50]

Originally Posted by Todd Madson

I never understood how people were cheating under Seti Classic. Can someone explain this?
...

There were several cheat methods used.

To begin with, Berkeley did not track what work units were issued to whom, and since they issued WUs to multiple people themselves, they also accepted a particular work unit back from anyone who sent it in.

So, person A on the team crunches a particular WU. He then passes the WU and the finished result to all the other people on the team. They all 'return' the same finished WU, though it was only crunched once by team member A. With 20 people on a team doing this, they can claim 20x the credit they otherwise could.

This method distorts credit, but doesn't mess up the science because the WU was crunched properly. It can be found if you sift through the server records, and find the same WUs consistently get returned by the same block of people.

Next, hacked clients. A hacked client accepts a WU, then immediately reports back that it's finished, and found nothing of interest in the WU. Zero processing, instant credit.

This cheat didn't work well later, because the later clients usually found gaussians and/or spikes/triplets in nearly every WU. An account that never found anything was cheating, same thing applies if the results didn't agree with the same WU when crunched by someone else. Again, Berkeley had to sift through the server records to catch this and ban accounts.

A variation was a hacked client that used simpler math. Something that found the more likely results, and somewhat agreed with the honest clients, but finished faster. Again, you had to catch it by comparing to the same WU when crunched honestly.

All the cheat methods could be caught with statistical analysis. The cheating was an issue because it became apparent the Berkeley people would rather do almost anything other than sift out cheaters. In one case, nothing happened until c|net wrote a story on the cheating.

The BOINC credit/verification system closes off all the old cheating methods. To my knowledge, no one has figured a good way to cheat the new system. So far anyway.

[Todd Madson]

Thanks for outlining that - I always presumed it was some
crazy thing like finding a way to make the result look like
10 or 20 on one PC.

Interesting too - that people were SO driven to be at a
higher number that they just tossed out the real purpose
for the project in the first place. Insanity.

Can you imagine the headlines: "Alien Signal Hoax,
Seti participant hoaxed system". Nobody would trust
them again.

So it's good they nixed this.