I got bored today and built a new 5.4.10 boinc client that increased the wetstone score about %50 higher and the drystone about %300 higher than the stock 5.4.9 boinc client on my laptop. To use these binaries you must have windows xp/2000 and a SSE2 capable processor. The binaries can be downloaded at https://boese.kicks-ass.net:7000. Enjoy!!!

---

In my opinion is this cheating.

---

I don't think there is a 5.4.10 boinc client,you should prove it!

---

<core_client_version>5.4.10</core_client_version>
See his Result ID. A compiler can label it whatever they choose.

NOTE: this is NOT official boinc software.

I was told the source code is readily available. That means if someone wants to improve or make changes, then they can compile a 'newer' version of the software. I'm sure there's some legalities out there talking about the distribution of the program. Dunno, I just keep plugging along. :)

---

To be honest, I dont care about anything but the science (can tell by my credits spread across so many projects and none particularly high). I did not change the code or config files illegally. All I did was change some of the compiler flags as well as optimizing the compiler flags for wxWidgets. From what I understand this is not strictly cheating. More than anything I was just bored and am a developer by trade so I felt like making a recent optimized build (got the source code from CVS today). And as pointed out this is not an official release and comes with no warranty. Just doing the internet open source thing and sharing. Thanks.

---

Using an optimized client is not illegal by means of law and the projects have no direct instructions about not using optimized clients.

But the optimized clients caused a lot trouble in several projects and all projects feel the need to look for a better credit system than the benchmark based one. For those who can read between the lines this is a clear point against those clients.

Otoh. (and afaik.) there are only 3 projects still affected very much by those clients, those three are Rosetta, QMC and TANPAKU, of which at least the first two announced to have a better credit system soon. I don't know about the plans at TANPAKU.

Besides those three, there are still a few benchmark based projects left, but those all have a higher quorum so the impact of a too high benchmark isn't that strong there, it usually shouldn't add more than 15%-20% to the granted credits in average for those clients.

Yeah the main reason I made the build is just for something to do though I have noticed that the boincmgr gui seems more responsive and to lag less than before with the new build (probably also partly due to also optimizing wxWidgets build). I also wanted a snapshot of the code more recent than 5.4.9 (build was off of CVS head on 7/29/06). The one bad thing is of couse the size of the dlls and exe went up some but they do perform better I feel. I will run it until at least the next official release but the whole edge to open source is being able to push the envelope especially as long as it doesnt affect the main priority which is the science.

---

Since the boinc manager makes no difference at all to the amount of science the project application does, optimising it has no real benefit, and certainly doesn't justify fiddling the credit calculations.


This result has generated 4 decoys, and requested 41 credit for it (i.e., 10 credits per decoy).
https://boinc.bakerlab.org/rosetta/result.php?resultid=30625026
FRA_t385_CASP7_hom001_5_t385_4_1sofA_IGNORE_THE_REST_288_1052_17_0

Whereas this identical job on another PC has generated 6 decoys for 37 credits (i.e., 6 credits per decoy).
https://boinc.bakerlab.org/rosetta/result.php?resultid=30627831
FRA_t385_CASP7_hom001_5_t385_4_1uvhA_IGNORE_THE_REST_148_1052_22_0

So your application causes your PC to overclaim by ~55% on the same actual work done. Why not just set it to overclaim by a million times?

---

CPU time 10961.421875
<core_client_version>5.5.0</core_client_version>
<stderr_txt>
# random seed: 2222475
# cpu_run_time_pref: 10800
# DONE :: 1 starting structures built 5 (nstruct) times
# This process generated 6 decoys from 6 attempts
# 0 starting pdbs were skipped
</stderr_txt>

Validate state Valid
Claimed credit 91.9309966512074
Granted credit 91.9309966512074

--> Still much work to do for this 5.4.10 boinc client. A really good CPU-optimized client like 5.5.0 should claim 92 instead 41 credits

---

Love, Michi

That's a whole new can of worms that I don't fancy opening right now! Perhaps we should just ignore the overclaiming for a couple of weeks until the new credit scheme goes in.

Is that a FRA_t385_CASP7_hom001_5_t385_4 workunit? It wouldn't be comparable otherwise.

---

That's a whole new can of worms that I don't fancy opening right now! Perhaps we should just ignore the overclaiming for a couple of weeks until the new credit scheme goes in.

Is that a FRA_t385_CASP7_hom001_5_t385_4 workunit? It wouldn't be comparable otherwise.

The results were taken from workunit FRA_t385_CASP7_hom001_5_t385_4_1bfrV_IGNORE_THE_REST_83_1052_16_0 but are comparable to all the other Rosetta WUs (approx. 30 Credits / hour CPU time for each core of an Athlon64 X2 3800+). Is it really overclaiming or just the fact how to play the game in the cruncher's premier league?

---

Love, Michi

I was comparing based on the number of decoys generated, which is why FRA_t385_CASP7_hom001_5_t385_4 is comparable but the others aren't (i.e., different proteins or different algorithms will result in different numbers of decoys being generated per unit time).

---

I was comparing based on the number of decoys generated, which is why FRA_t385_CASP7_hom001_5_t385_4 is comparable but the others aren't (i.e., different proteins or different algorithms will result in different numbers of decoys being generated per unit time).

Fogive me if I am wrong (and we are going off the original topic, sorry again) but
You are comparing the number of decoys, what is to say the a decoy didn't take longer to create in on than the other ?
Isn't that the whole point of a benchmark like this, to overcome the fact that the times are variable in this sort of work. A decoy (afaik) does not have a fixed length of time/calculations, but it's finished when it's finished (as you do mention above) but even from a different random number start as well ?

Though we know full well it'll be claiming higher credit then 5.4.9, credit/hr is simpily both bench values added togother * 2.083

Back to the original topic
Though I don't remember a 5.4.10 client being put together since the development brach (all versions after 5.4.9) are under 5.5.x
Don't know why people give you the negative mark, a least you managed to compile it ;-)
(Though if you compiled it with MS VC++ 2005 Express, a how to would be nice, I can never seem to get it to work, you may also want to pop onto the dev list and help them with it, especially the boincmanager)
v5.5.9 is the latest release development version.

---

Team mauisun.org

I just noticed that in the case of the two proteins that were presented , the crunching was stoped by the watchdog. So the time where the watchdog stopped, the place in the molecule where each machine started among other things may have affected the number of models cretaed and the amount of credits claimed.

Can I ask you all a favor?

Why dont we concentrate on what it is important'now: CASP 7. It is less than 9 days till the end. Let's concentrate in helping Baker Lab finish their predictions.

Once CASP 7 ends and the developers can take a breathing break, they will present/unveil the new credit system. Then we ask the questions. Trying to guess what is comming and worst trying to find the faults of something we dont know yet is an excercise in futility; it is a waste of time.

When one does a cluster anaysis of the models done by the computers vs the credits claimed, one sees very defined patterns that can allow a fair allotment of credits per model created in similar work units. Let's say the analysis has been done with a large sample .

But, as I said before, right now it is more important for me to get sloth and Boinc and Rosetta to work and crunch without errors than to start a discusion on something that can wait.

So lets crunch...August 8 is sooner than we think.

I tried 5.5.9 this afternoon but I was unable to download a single WU. Back to 5.4.9.

As for the client version 5.4.10 was just what popped up after I built the CVS head so I ran with it. The release config for the winbuild is actually broke right now in CVS and required some fiddling. The new build has been running stable on my machine since I built it. Based on how much more stable the CVS head is running right now compared to a month ago I have a feeling a new official client is coming soon. As for the %100000 question sure you can cheat much worse but simply streamlining a build is hardly cheating. I mean why should I have to use the lowest common denominator slow build because of the few people still running 486s out there. Why dont we all run 386 linux kernels also :P.

---

The BOINC client has nothing, absolutely nothing, to do with science, better cruching, speed of crunching or whatever.
It's only for the administration of the science applications, and for credit claims.
From a scientific POV there is no need whatsoever for an 'optimized' client, the science is simply not affected.

So the only reason for running an 'optimized' client is to get more credit than supposed (or lack of knowledge of the process).

---

Grüße vom Sänger

Doh! just checked my alias for cvs command and realized this is not the CVS head after all but the stable branch. That would explain why it is 5.4 instead of 5.5. Oh well probably better anyway especially to avoid any problems with CASP. I at least didnt change any of the source like most of the other optimized builds. Yea h thats it I was just getting the latest bug fixes out early and my compiler accidentally added flags like /Ox /arch:SSE2, etc. Thats the ticket :).

---

Jose, I just wish people wouldn't jump on someone that actually tries to compile the program themselves. It is open source abnd people are allowed to do it. If they wanted to stop it they could build in restictions (certificates blahdey blah blah etc..). But don't have a go a someone giving things a try.
Yes we know the credit is a mess, yes we know rosetta are bringing something new. And I at least know the 8th is very sson (I ahve my oral presentation/viva for my MSc then :-s ).

Now you say BoincManager runs better :-) I'll give that a go, the current one is sluggish and BoincViews to much for a single computer.
Is that just a compile time optimisation on wxWidgets?

As for getting on with CASP7, what has that got to do with this ? This in know way slows CASP down. It's just a boinc client.
(and P.S. I am, two computers just went full time for a week, one of them a new one)

---

Team mauisun.org