Today evening i'am out of the race, because it doesn't make any sense to run a challenge against cheater clients.
P4 machines are earning more points then latest c2d, quads or Xeons. *crazy*

No thanks to this!

and now we discuss to stop challenging !!
we are challenging with fair play against AF last one
we are not challenging without fair play against SUSA this time
my opinion, others will follow ...

and now we discuss to stop challenging !!
we are challenging with fair play against AF last one
we are not challenging without fair play against SUSA this time
my opinion, others will follow ...

What's the problem !! I don't understand what is fair-play or not ?? Is somebody cheating here ?? I don't like cheating so tell more or nothing please .

I will state unequivocally that Seti.USA has *NEVER* cheated on any challenge. I will be notifying Team Leadership.

I request Primegrid Admins remove those unsubstantiated accusations from the boards until they can back them up (which they can't)

I will state unequivocally that Seti.USA has *NEVER* cheated on any challenge. I will be notifying Team Leadership.

I request Primegrid Admins remove those unsubstantiated accusations from the boards until they can back them up (which they can't)

I also run one of these 'GenuineIntel Intel(R) Pentium(R) D CPU 3.00GHz' that ticks a WU over in around 19000 seconds (whilst having some other activity going on). That's no different than what you call a 'cheater'. Am I a cheater too? I emphatically am not, as will be proved by the validator when the double-check comes in.

And when the "cheater"-computers are all overclocked??? You can´t see the real cpu-clock in Boinc

CPU type GenuineIntel
Intel(R) Pentium(R) D CPU 3.00GHz [EM64T Family 15 Model 6 Stepping 2]
Number of CPUs 2
Operating System Microsoft Windows XP Professional x64 Edition
, Service Pack 2, (05.02.3790.00)
Memory 1013.87 MB
Cache 488.28 KB
Measured floating point speed 1357.12 million ops/sec
Measured integer speed 3945.19 million ops/sec

16,486.23 114.09 170.98
18,615.27 128.82 115.87

Measured floating point speed 1357.12 million ops/sec
Measured integer speed 3945.19 million ops/sec

I see nothing odd about the benchmark speeds on this client.. just what are you saying. In fact the benchmarks seem a bit low for a Pent D at 3.0 Benchmarks at 1357 ???? should not that benchmark be 1500?

And, Just how is BOINC 5.10.30 a "cheater" client - isnt this a standard BOINC release? a bit outdated, but still a Berkeley released version.

We discussed this enough and you are aware what you are doing.

Examples from none cheating 5.10.45 client on 64bit WinXP

E5420 Xeon Standard @2500
12,055.44 68.50 81.49
12,032.78 68.37 82.58


And if you are thinking that a P4 has more power then dream further
Nothing else to say because we all know SUSA is untouchable.

@FrankHagen
16,036.73 36.66 47.36 yours and make a comparison to the earlier posted results from me. ;)
They are claiming 300% more then you with a related CPU

And also from my point I know what you are doing.

@FrankHagen
16,036.73 36.66 47.36 yours and make a comparison to the earlier posted results from me.

I will state unequivocally that Seti.USA has *NEVER* cheated on any challenge. I will be notifying Team Leadership.

I request Primegrid Admins remove those unsubstantiated accusations from the boards until they can back them up (which they can't)

Used is your known cheater client 5.10.30
Your cheaters are over granting newest 45nm CPUs with 200%
Nothing else to say, bye i am crunching for another projekt

So kindly base your accusations on TRUE facts, not baseless ones.

sorry if i step in again - it's only my 5 cents:

TGIF - someONE (over here) will have to take his time 'till monday and see his shrink then.

frank.

It's all good Frank...I've had my say :)

It's all good Frank...I've had my say :)

so let's chill now and enjoy the party to the full!

happy crunching..

frank.

To stop all suspicious things can you please remove all credits ( like other project does ) from the owner ( cheater ) of this host : http://www.primegrid.com/results.php?hostid=71458

It is anonymous so no team accused but this guys is a cheater !!!

To stop all suspicious things can you please remove all credits ( like other project does ) from the owner ( cheater ) of this host : http://www.primegrid.com/results.php?hostid=71458

It is anonymous so no team accused but this guys is a cheater !!!

Just read carefully the integer speed and the claimed credit

Created 17 Jul 2008 17:44:40 UTC
Total Credit 9,751
Avg. credit 287.35
CPU type GenuineIntel
Intel(R) Core(TM)2 Duo CPU E4500 @ 2.20GHz [x86 Family 6 Model 15 Stepping 13]
Number of CPUs 2
Operating System Microsoft Windows XP
Home Edition, Service Pack 3, (05.01.2600.00)
Memory 2047.11 MB
Cache 488.28 KB
Measured floating point speed 2820.48 million ops/sec
Measured integer speed 149126.92 million ops/sec

Claimed Credit : 1,110.15

regards

Agreed, that is excessive -

I don't know how the credit system works for this project, but it appears that both WU's in the quorum received the same amount of BOINC credit,

http://www.primegrid.com/workunit.php?wuid=38075824


but this still does not affect the race credits.

EXTREM CHEATER
http://www.primegrid.com/show_host_detail.php?hostid=71458
http://www.primegrid.com/workunit.php?wuid=38082298

Just read carefully the integer speed and the claimed credit

Created 17 Jul 2008 17:44:40 UTC
Total Credit 9,751
Avg. credit 287.35
CPU type GenuineIntel
Intel(R) Core(TM)2 Duo CPU E4500 @ 2.20GHz [x86 Family 6 Model 15 Stepping 13]
Number of CPUs 2
Operating System Microsoft Windows XP
Home Edition, Service Pack 3, (05.01.2600.00)
Memory 2047.11 MB
Cache 488.28 KB
Measured floating point speed 2820.48 million ops/sec
Measured integer speed 149126.92 million ops/sec

Claimed Credit : 1,110.15

regards

I'm right there with you, [SG]marodeur6 and I'm also very disappointed!

Hi Rytis!

____________

This is really disgusting, the whole gang of 6.1.0 Cheaters gathered again. I would strongly recommend not only to decrease future misuse by this limit. They have already profited enough by there cheatery.
Please set a sign and block the app. 6.1.0 which most of the cheaters use. The boinc software is able to exclude cheating clients. So show them, that you do not want further cheating. Noone uses 6.1.0 and is unaware that it is cheating!

Throw them out! Now!

For those who don't know what the 6.1.0 does:
It was developed to balance the credit loss for crunchers who used crunch3rs seti application. To do that it overclaims credits. In the case of seti this is on the border of being tolerable (since someone just decides to earn more credits than others, not very fair in itself) but using this overclaiming creditmachine at any other project without using an appropiate application is only one thing: mere naked uncovered cheating)

@blurf, it might be true that noone of your team was cheating here at the challenge, but this is only a question of definition. Obviously this above mentioned client is used by members of your team, probably also by some others. So for me there are a lot of cheaters within your team. But if you deny overclaiming to be cheating, than you might be right. Also as this does not concern the racecredits, you might further your argumentation of noone cheating with the race.

But doesn't this misplaced overcleaiming spoil the whole credit-system, doesn't it spoil your team's number 1?
____________
Dragons can fly, because they don't fit into pirate ships!

But doesn't this misplaced overclaiming spoil the whole credit-system, doesn't it spoil your team's number 1?

A thread to discuss BOINC crediting and benchmarking.

Counting cobblestones for challenge with non-fixed credits would be the worst solution. For one WU you get 80 cobblestones, for the next one (with the same runtime) you get 120, because your wingman has a slow AMD processor, for the third one (again, with the same or even a shorter runtime) you get 400 because your wingman has inflated benchmarks...

Any parity in cross-project credits is a myth.
This discussion is not about cross-project parity, but about the way credit is granted in LLR subprojects...

I will say again, claiming amounts have NOTHING to do with the challenge scores. Let me quote the project again:

There is no correlation between Challenge Points and BOINC credit (cobblestones)! There is ABSOLUTELY NO WAY to cheat Challenge Points.

There is a relatively easy fix to inflated benchmarks that could be implemented by any project (though it would require some extra programming). Benchmarks are natually inexact because:

1. There is variability between machines of different speeds.

and (very importantly)

2. There is variability within a machine since benchmarks vary across time due to variability is processor load at the time of benchmarks.

Thus, this is a stochastic process that should be considered using a statistical frame of mind. If we assume that this variability is spread out through a fair (read random, stochastic here) process, then the distribution of the benchamrks scores should be roughly normally distributed. This would ideally be applied at the processor (and perhaps OS) level, but more generally could be applied across all types of machines using a single benchmark distribution (or at least one each for integer and floating point benchmarks).

Thus, any excessive benchmarks (say outside the 99.9% confidence bound--high and low) could be identified. Therefore, the LLR scoring algorithim here could be modified as follows:

1. benchmarks from both machines within 99.9% bound, use current scoring process

2. one machine within 99.9% bound, one machine outside of 99.9% bound--assign scoring claimed by client within the bound

3. both machines outside of 99.9% bound, assign to an additional machine until a result from benchmarks within bounds is returned and assign that score to all machines.

While this does not solve all the issues noted above, it does tend to average over them and does eliminate extreme values.

1. benchmarks from both machines within 99.9% bound, use current scoring process

2. one machine within 99.9% bound, one machine outside of 99.9% bound--assign scoring claimed by client within the bound

3. both machines outside of 99.9% bound, assign to an additional machine until a result from benchmarks within bounds is returned and assign that score to all machines.

While this does not solve all the issues noted above, it does tend to average over them and does eliminate extreme values.

I'd say 95% is still too much. But yes, it would probably be a pretty good system; I'll gather some statistics to see if it is doable.

just use the challenge formula to grant llr credit. that is multiplying it with a constant factor to match current credits per test.

I updated my post, my idea is not to resend anything, but instead use adjusted benchmark values.

You do not need any modified BOINC client to inflate benchmarks. Anyone can do that with any stock client downloaded directly from Berkeley. Trying to ban over-claiming by banning certain versions of clients is a fool's game.

I think it's wasted effort banning by version number. It will block some valid clients, and still allow messing with benchmarks. I think effort would be better used coming up with a real solution. Simple as that.

I'm simply not going to look at the benchmark values provided by the client when assigning the credit, I'm going to look at a BOINC-wide median benchmark value for that specific host type.

Actually, a year or two ago, I received non personal data for all hosts attached to a project of this size to do an analysis to see if an "median benchmark" could be established for all known hosts. Establishing a "median benchmark" and a filter to flag hosts which deviated more than 50% (to allow for serious overclockers) would be effective, as most "cheating" boinc clients increased benchmarks significantly more than that. The trouble come in when you realize just how many variations of hosts and OSes are really out there. Some only had ONE listing, meaning the median would be itself. If that one cheated, then noone would know. Also, there's nothing to stop someone from changing the data that Boinc extracts. Then someone could label his Celeron 500 as a much higher scoring machine. Heck, they could probably get it to report their puter as "the good ship lollypop".

I scrapped the idea when I realized just how much of this would require LOTS of man hours for EACH project. At that point it seemed less likely to be adopted. If Rytis can do it, I say let him.
I could/might reboot my machines to windows after the challenge and look it up.

[edit] thinking back, I did this before Boinc started using MSVC 2005. When Boinc switched from MSVC2003 to 2005, it resulted in markedly higher benchmarks for everyone, so my data would reflect hosts using MSVC2003.

[2nd edit] banning one or two versions of those "cheating" clients would only be effective on those numbered clients. There are now nearly a dozen freely available ones, and nothing to stop anyone with a compiler from coming up with his/her own numbers, or even just naming it the same as an official version and slipping past this kind of snare.

banning one or two versions of those "cheating" clients would only be effective on those numbered clients. There are now nearly a dozen freely available ones, and nothing to stop anyone with a compiler from coming up with his/her own numbers, or even just naming it the same as an official version and slipping past this kind of snare.

I've looked up the old data. It was extracted on Dec 5th 2006. It contains 4955 active(having contacted the server in the last 7 days) hosts from that project(that number started at much higher until I pared out as many duplicates as I could fairly judge to be duplicates). It contains 659 distinctly different "Model" and "OS" (Win or Linux) combinations. 265 of them are "Unique" entries. Meaning they're ONE of a kind and thus a Median, Mean, or Avg is futile. An addition 114 have only ONE like it and I suspect some of these are really unique but are showing up as to due to the detach/reattach process.

THe establishment of a "mean" would be effective at finding the "high claimers" from many models though which are in frequent use. Is that of any use? Wouldn't there still be a few who "modified" something or other to "game" the system. Is it worth the effort of the project admins/scientist/engineers???

here's a Sample from the P4 3.40 Ghz using Windows.
The Average Claimed credit/hour is 7.49, the median 6.83, there were 58 hosts, the High claim/hour was 12.17, and the lowest claim was 2.65

Here's the P4 3.0 Ghz, There's 356 samples, The Avg claim/hour is 6.26, the median 6.00, the High claim/hr 10.54 and the lowest 1.58

EEEK, it's been so LONG since I've done this. I forgot, the last two charts have had all 5.5.0 hosts removed for some long forgotten reason. Back to the databases to show them with 5.5.0.

Another eek. For me to show those charts, I'd have to rework from the original data. Meaning many hours that I'm not willing to put in, so I just looked up the high/low claim for those 5.5.0 host and P4 3.0 ghz and windows. The low was 13.42 credit/hour, the high was 17.03. If Plotted, in the chart below, you can see how glaringly high it would be, and that a filter set to 50% higher than median would have filtered them out. Leaving the overclockers to do their best...undetered.

I've just spent all the time from my last post to now getting this chart ready for just the P4 3.0Ghz computers using Windows. The total hosts now is 375.

The X axis shows Boinc Verions, but is sorted by Claimed credit/hour ascending. (like the earlier ones).



As you all can see, there is a distinct difference between claims of normal hosts, a slight jump for what I'm assuming is Overclocking, and then the "third party" boinc clients.

Also, there were only 19 5.5.0 clients out of 375 hosts, so not a high percentage of users were using it.

Similar results to this probably exist with newer data.

Hope this helps Rytis.

I've just spent all the time from my last post to now getting this chart ready for just the P4 3.0Ghz computers using Windows.

And how are things going? Have you found a favorite strategy for the future?

does anyone has by the way a statistics that shows how much % of work is delivered by the clients 6.1.0, 5.5.0, and 5.9.0?

and very off-topic @ rytis: Is chess still alive? Were you not somehow the one who was sometimes kicking the server?
____________
Dragons can fly, because they don't fit into pirate ships!

We are going to try median BOINC-wide benchmark values for specific kind of CPU type

but it's up to Willy from BOINCstats to make a table (it will probably happen this weekend).

and very off-topic @ rytis: Is chess still alive? Were you not somehow the one who was sometimes kicking the server?

and very off-topic @ rytis: Is chess still alive? Were you not somehow the one who was sometimes kicking the server?

Quick off-topic note: I'm working with Joerg on Chess960. Hoping it'll be back up in September.