I hate cheating. A lot. I spend a lot of time tweaking my computers for maximum rosetta performance - and even more time staring at my stats. So, as far as I'm concerned, cheating is not just annoying, it actually ruins the fun completely.

Following a recent thread on how WU redundancy might be one approach to a potential solution, I decided to go and dig in the (fantastic, BTW) boincstats to spot potential cheaters.

Now I'll let you be the judge - here's a link that sorts all the hosts by average credit and includes the reported processors: here

As you can see, we have users in there with P4 3ghz HT benching at 1735 / 6562. Funny enough, most CPUs of that type bench around 1320/1590. Not that anyone would take offense, as those users are named 'n/a' or 'anonymous'.

The mind boggles.

---

Team CFVault.com
http://www.cfvault.com

1,229 credits in one day with a 3GHz P4? (id 37936) Right...

EDIT:: I have one WU for which I got 200+ credits. It took 31 hours. So it's _remotely_possible_ that one day's stats could be this far off. But a consistent RAC of 770+ for a single unexceptional computer, with some days being well over 900, sounds, um, a bit high.

---

Interesting, all though maybe I missed it but I don't see where the Benchmark scores actually show up ... ???

Interesting, all though maybe I missed it but I don't see where the Benchmark scores actually show up ... ???

Hi Poorboy, you'll have to write down the Host id for the suspicious machine, then go to https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=XXXXX (and replace the hostid at the end of the url with the right one).

---

Team CFVault.com
http://www.cfvault.com

P4 3ghz HT benching at 1735 / 6562 is coreclient 468sse2.
Hi Have P4 3ghz HT benching 3376 / 5119 coreclient 472 sse2.
Bench is very important for credit.
Time crunching for 472/468 is equal.
Sorry for my english to write for translator.

---

P4 3ghz HT benching at 1735 / 6562 is coreclient 468sse2.
Hi Have P4 3ghz HT benching 3376 / 5119 coreclient 472 sse2.

Can anyone confirm/refute that Boinc 472 gives out hugely inflated benchmark values?

---

Team CFVault.com
http://www.cfvault.com

P4 3ghz HT benching at 1735 / 6562 is coreclient 468sse2.
Hi Have P4 3ghz HT benching 3376 / 5119 coreclient 472 sse2.

Can anyone confirm/refute that Boinc 472 gives out hugely inflated benchmark values?

Not the official 4.72 client, only an optimized version (boinc.exe)

---

Hi have changed coreclient 472sse2 optimized
to coreclient 468sse2 optimized new bench 2148.32/8072.85.

---

Hi have changed coreclient 472sse2 optimized
to coreclient 468sse2 optimized new bench 2148.32/8072.85.

That's 'optimized' allright :-)

---

Team CFVault.com
http://www.cfvault.com

I'd like to know what the admins are planning to do about this, as the cheating - conscious or not - is blatant.

---

Team CFVault.com
http://www.cfvault.com

personaly i hate people who would think of cheating but its only numbers

http://www.boincstats.com/stats/boinc_host_stats.php?pr=bo&st=0&to=100&userid=5e6c75b7af8104f0835d40b545e944b0

this is the computers im running i dont run anything special and infact i a guessing i actualy may be getting lower points than others. I still dont understand this benchmark stuff your talkin about though.

can you tell me how to check the benchmark stuff i may have 4 computers running but i dont know everything about them.

if there is a point system or such someone will try to cheat.. my bigest concern is, if anything that they do when they cheat messes up the results. Since in the end the results are actualy what matters. I am sure you will find that MOST if not ALL of the people comming here from FaD hate cheating also but love tweeking there computers and such to get the most from crunching

---

I'd like to know what the admins are planning to do about this, as the cheating - conscious or not - is blatant.

Not important crunching for cheating or not conscious or not.
Important is Project.
For information you see http://boinc.berkeley.edu/download_other.php

---

I'd like to know what the admins are planning to do about this, as the cheating - conscious or not - is blatant.

Not important crunching for cheating or not conscious or not.
Important is Project.
For information you see http://boinc.berkeley.edu/download_other.php

In that case, why do you run an optimized client that artificially gives you a higher score? The whole point of a benchmark is that it's only useful if everyone runs the exact same version. That's the only way the benchmark can do its job and accurately reporty your system's performance.

---

In that case, why do you run an optimized client that artificially gives you a higher score? The whole point of a benchmark is that it's only useful if everyone runs the exact same version. That's the only way the benchmark can do its job and accurately reporty your system's performance.

It does make sense when running optimalized application as well (case of SETI).

I have never seen a benchmark that "accurately reporty your system's performance." It's only a model and any model as such simplify the nature of the problem. To be honest, it's many years since I put stress to benchmarks. Today, I rather look at 'real' speed of processing.

Examples of limitation on BOINC benchmark are various and - for sure - had been already made somewhere so here goes just top of my head:
- size of the CPU cache is more or less important for different project; cache size has lower of different importance for benchmark than for live application
- some projects 'prefer' AMD, others Intel - due to different compilers and switches and BOINC benchmark ignores that.
- BOINC benchmark doesn't not account GPU performance. When screensaver or graphics in each project is on, this definitely has some impact on speed processing (effeciency) - and different due to GPU used, drivers used etc. Users with superb VGA will computer faster than those with old on-board VGA with shared memory.
- BOINC benchmark doesn't account 'loss' of processing while apps switching of those people, who do no leave apps in memory (preempting). People with preemting on should be more effective (if they have enought memory) or be less effecite (when low on memory, more projects attached and swapping takes place).
- some people degragments their HDs, some don't - this might have considerable impact on CPDN speed for example. Speed of HD is ignored on BOINC benchmark. More - it's ignored where BOINC is run on: it does make difference when you run BOINC from a LAN behidn SW firewall or from a local RamDrive.
- with BOINC, it is unable to tell where the BOINC data are, nor where the BOINC temp dir is. Benchmark ignores if system, swap and BOINC are on the same disk/partition.
- BOINC benchmark is not the same for all OSes

So, even when everyone would be running same BOINC, many parameters of system are not measured and/or ignored. But all of them have some impact on overall BOINC speed. Hmm, only benchmark...

How can one assure that everyone is running same BOINC? Major version gives different benchmark and - I believe - there are more computers still running 4.x than computers running any kind of optimalized version.

nb: some people are running tru'x BOINC core 5.321 spoofed as 4.531 since they were able running BOINC Manager 5.x (hence suspend function and many more) with projects not ready for BOINC 5.x genration.

---

Since in the end the results are actualy what matters. I am sure you will find that MOST if not ALL of the people comming here from FaD hate cheating also but love tweeking there computers and such to get the most from crunching

There are two, very distinct, issues at play here:
- Optimized clients for platforms where the regular benchmarks are naturally low, trying to 'even the playing field' - and fine by me. I don't think anyone as any problems with that.
- Cheaters, AKA people who fake their numbers to get higher up in the credit ladder. I was hoping everyone would have a problem with that.

But these are 2 different things. Arguing that somehow people who are against cheating are against optimized clients - is a logical fallacy - a strawman if you will.

I'm was quite taken aback to see so many people stating that 'it's just numbers'. Unless you have been living under a rock for the past 5 years, you should know that what really drives DC projects are the stats. Look at Free-DC, they even state in their profile 'why should you join our team? We have the best stats'. Team Ars Technica and Team Hard OCP, to only name 2, always have competitions that bring a lot of credit to DC projects. If you never heard of those people, you should probably check the stats. Free-DC, for example, now brings 4 times more units than the behemoth Housing and Food Services itself.

That those people are stats-driven should start ringing a bell!

I've seen people buying new machines and specing them to perfection with the goal of crunching better in mind. I've seen people not hesitating one minute to get older machine run 24/7 and drive their electricity bill up just to get that one more credit. It's downright belittling to say 'these are just number' with a dismissive attitude. My WU are not worth less than yours just because they are stats-driven.

---

Team CFVault.com
http://www.cfvault.com

There are two, very distinct, issues at play here:

There's actually a third, that's a little bit grey.
1.) Optimized client + Optimized science app (tweaking for a level playing field)
2.) Optimized client + Standard science app (Grey cheating)
3.) Benchmark fiddling (Blatant cheating)

Number 2 works on the following premise:
"I know that my scores are going to be inflated beyond the norm, but if anyone accuses me, I'll just claim that this client is attached to another project which runs an optimized science app, therefore it's not my fault and I'm not cheating."
or
"I know I'm cheating, but if I'm found out, I'll just claim ignorance."
or
"Tough luck. Don't complain. If you don't like it, submit and join us."
or
...I think you get the idea...

Of particular concern to this project's participants is number 3. If I wanted to, I could make a 200Mhz system claim 10000 credits (or, "insert your wish here") per result, but I choose not to because I abhor cheating. The solution to this problem that is easiest for the project to implement is a quorum system on returned results.

It's downright belittling to say 'these are just number' with a dismissive attitude.

The way that I read it, I believe that the individual was stating their own point of view regarding cheating to inflate numbers, and as it happens, they are not alone in that view.

There is no argument that incentive drives competition in some spheres, but believe it or not, there are some participants who don't actually care for the numbers, and that's their prerogative.

The way I see it, your argument may be more valid if it were directed at a project owner who said "My project doesn't need credits, they're just numbers...". Participants participate for their own reasons.

My WU are not worth less than yours just because they are stats-driven.

I'm not entirely sure where the accusation is that the comment above is directed at defending. [confused]

---

My WU are not worth less than yours just because they are stats-driven.

I'm not entirely sure where the accusation is that the comment above is directed at defending. [confused]

It wasn't directly at anyone in particular :-) I was just refering to, in general, people who look down on stats and therefore automatically oppose anything that prevents cheating since it may have side-effects. I don't think that's healthy for the project, but as you said it's their prerogative.

edit: happy hundredth post :-0 Yes there are stats to be had everywhere :)

---

Team CFVault.com
http://www.cfvault.com

There's actually a third, that's a little bit grey.
1.) Optimized client + Optimized science app (tweaking for a level playing field)
2.) Optimized client + Standard science app (Grey cheating)
3.) Benchmark fiddling (Blatant cheating)
===========
It is my understanding that an Optimized Client does only 1 thing, that is to inflate the Benchmarks for that Particular Computer.

An optimized Client bundled with an Optimized Application does 2 things, inflate the Benchmarks plus speed up the processing of the WU.

With that understanding I don't really see how it is any different than Blatant Cheating is. If you use an Optimized Client your getting more Credits than you were intended to & if you use the Optimized Client with the Optimized Application now all your doing is getting more Credit and getting it faster ...

I know the standard reply is that it's for the Science, that's a bunch of BS as far as I'm concerned. All it boils down to is getting more Credit in the end. Unless everybody runs the same Client & Application @ a Project then somebody's Cheating...IMO

Another Reply is "Well everybody can install them, or Their just to Lazy" ... That's not the necessarily the case in either instance. Some people are just Computer illiterates and maybe can't figure out how to do it. And some people just don't care or just don't know about the Optimized Clients & Applications because they are not that Fanatical about their Credits and spend their every waking moment in the Forums to find out about them.

Jeesh, we already get a new Client from Berkeley every 2-4 days that's hard enough to keep on top of, then we are supposed to keep up with all the Optimization thats going on too ...

I also keep my mouth shut anymore when I run across something that seems quit fishy because I got in in trouble once before for saying something. Everybody sided with the person that was cheating or at the least fiddling with the Credits & I was made out to be the villain.

Right now I know of a serious & provable case of cheating going on at one of the Projects, but I'll be darn if I'm going to say anything about it. Let the chips fall where they may is the Motto I go by anymore ...

If you use an Optimized Client your getting more Credits than you were intended to & if you use the Optimized Client with the Optimized Application now all your doing is getting more Credit and getting it faster ...

Ad 1)
No quite so. If one runs Optimalized Client - that gives let's say 2x benchamrk than 'standard' (e.g. average 5.x version) and you run optimalized apps like trux's SETI apps that runs 3x or more faster (hence 30% of original time), you actually get credit as follows: bechmark * time-spend-on-WU.
well, according to my math - and I could be easily wrong - that gives * 2 * .30 which is 0.6 (or 60%) of originally claimed credit (by the standard client+standard application). You do 3x more science jobs but claim only 2x more credit. Is that cheating?
If one gets 3x more benchmark and 'only' 2x faster application, this would results in 150% of claimed credit comparing to standard client/apps.
Sure - this only apply when there is no quorum mechanism involved.

Ad 2) on ideal but what one can do when some projects have optimalized apps, some only claim to release the source code, some are not intending to do so (for example due to copyright restrictions).

Ad 3) that's a bad case and nobody from regular users like to see that. I think this could be easily filtred out in databaze as low correlation of CPU-benchmark comparing with rest of machines (or a typical machine).

---

Any way you look at it if your using an Optimized Client and an Optimized Application your claiming more Credit & claiming it faster than you would if you were not using the Optimized Client & Application over somebody that's not using the Optimized Client & Application.

Abettedly your claiming less Credit because of the faster processing of the WU's, but your still Claiming more Credit than you should because of the Optimized Client no matter what speed you process the WU's at.

So in the end it all comes down to Blatant Fiddling with the Benchmarks because some people feel they are not getting what the Credits they deserve whether rightly so or not.