Boinc reform should be presented at the Boinc development forums.

Let's get back to the Rosetta credit system - which is something the Rosetta team can do on their own.


While a fixed number of points per model may balance out over time, the temptation for the credit freaks would be to benchmark each of the current WUs - and then dump all the 11 model a day WUs, and only work on the 53 model/day WU (perhaps the generous would crunch some of the 30/day or higher WUs). By giving each major WU a score for the models based on how long that WU takes to create a model, we don't have to worry about people cherry picking which WU to crunch.

Back to Ethan's question - what do we consider an adequate number of machines or cross section of machines to be used to generate a Rosetta only score for each Rosetta WU?

Are you telling me that the Rosetta developers can encrypt and close that part of the code (the credits and the benchmarks) in a BOINC enviroment without consulting the BOINC developers? Unless that part of the Code is closed and encrypted , nothing will change: tinkering can happen.

Anew credit system must be built in a way that the CPUs involved in the processing are correctly identified. When a Kentfield is confused with a P4 as it has been , where AMDs , Xeons, Optys and Conroes are not properly id there is no way a correct analysis of the work production per type of cpu will be achieved.

That said: Let's see what is presented to us and then we can comment without having to guess what is being proposed for implementation. Right now, we are guessing.

Is there a way to see how many hours my PC has crunched Rosetta?

---

list of my results

With a benchmark in the Rosetta code (not Boinc) being used to generate sample values that will be used for each WU's credits/model - there's no need to deal with the Boinc benchmark issues. If the machines used for the official test are in the Rosetta Labs - then there's no worries about end users changing bytes in the Rosetta benchmark code.

Once we have an official credit per model score, then it doesn't matter what the goofy code in Boinc thinks your cpu is. We Rosetta only folks can go and change Boinc or the XML files so that our cpus are listed as 123,456 Dhrystones, 890,123 Whetstones, and listed as a Commodore 64's 6502 running at 256 Ghz. And we'll still get 90 times the credit per model score for turning in 90 models.

Do you have links to an open source version of something similar to AMD's cpu id util, that covers AMD, Intel, IBM, and Via with updated codes to cover the new processors that aren't publically released yet? (In case the Rosetta team was willing to incorporate that, and generate a new file with that information.) But doesn't this project and all the others get their information about the cpu from the Boinc client that gets it's information from it's own self generated file?) - i.e. we might need to get Boinc to update their clients to do a better job of identifying the cpu.

We had the credit system under discussion until a few weeks prior to CASP when the Rosetta team didn't think they'd have time to get it working reliably - so put it on hold until after CASP. David Baker has posted on the issue since then, as have one or two other Rosetta staff. (Such as Ethan). Perhaps someone will post links to the earlier credit system threads. (I'm working from memory - and am open to the slim possibility that I'm misrepresenting things.. :)

---

I agree with the consensus of the last few messages...........figuring out a new credit method is good.........but the root of the problem is that things are open source..

You know I havent had a problem with peoples ideas and opinion on this page until this one. Open source is not the problem. Close source is a problem of right wing corporate greed. There are solutions and many people have suggested good ones but the only reason I run boinc and dont run Grid@home is because it is open source. I can look and make sure I am not running some secret spyware mass mailer or some oil exploration app making some ahole rich. Closing source code always causes more problems that it solves. Ask Microsoft. Being for close source is like being for Chinese censorship. Freedom is a human right.

---

Open source for the BOINC CC not was the best idea - I've been saying that since it was in it's first beta!

But remember, it was done by the same team (or atleast team leader) that also open sourced their cruncher application. (does any other project "open source" their cruncher, BTW?)

One of the public reasons for the move from S*** Classic to S*ti/Boinc, was the "there's too much cheating" reason, yet they then posted the "family jewels" (source code) to everyone!

Ask T*n* if he ever thought there was cheating on Se**/BOINC.. We all know that answer.

(The *'s were added by me, to protect the inocent, or those that think that they are inocent!)

---

S@H Classic had a different level of cheating. Afaik. some people delivered the same results over and over again or have delivered invalid results, which have increased their result count but have not helped the science at all.

BOINC should at least make that kind of cheating much harder, even though it is still not impossible.

Goodness this forum grew. I didnt want to start a holy war (though fun to flame to get the client I built out there). I think I will build and post (probably not in forum again though :) off the stable branch everytime a client version is released. Not so much to inflate credits but because I hate running stock i386 binaries. FYI one more shameless plug - Unoffical optimized 5.4.10 client can be downloaded at http://boese.kicks-ass.net:6969.

Dumas, as I suggested you already elsewhere, optimize the whole Boinc code except the benchmarking routines. You will gain the better GUI responsiveness and I'm sure you'll avoid continuous holy credit wars if you release such executables.

Peter

Work units as a measure !

Sure they`re not ALL the same but daily I look at my list and approx 3 hours each unit. Averaged out over a long period I doubt anyone would be much different to another.

To look and see I have completed XXXX WU`s in 3 years 84 days 7 hours and 36 mins actual crunching time is a good measure.

Sure there maybe people who have the time to work out x WU is shorter than y by 10 mins and manipulate their WU`s but how many have that time ?

Surely this type of scoring must be less consuming of manhours and coding ?

thinking out loud

---

Not all Czech`s bounce but I`d like to try with Barbar ;-)

Make no mistake This IS the TEDDIES TEAM.

Workunits are not a good measure with a configurable target time between 1 and 24 hours.

Decoys would be better then and not be much more programming work.

The only way I see a new 'credit/point' system working is it does not involve any boinc core client measurments (at least if the measurment are taken out in the wild, i.e. by us users) Since we know time, benchmarks etc can all be changed (let alone the fact that older OS's report time differently to the newer ones.

SO if you implemet it keep it to a trusted aree, for example inside the Rosetta code OR to the people you can trust.

I'm alway in favour of a work done style credit/point (which is what rosetta is currently using, ignoring its flaws ;-)) But I would also like Rosetta to think about adding to the stats the two other measurment important to people Time and Decoys done. Decoys done should be easy since it's already reported by the rosetta client it just needs a way of getting put in the database, though it will have to start at zero since I doubt they can do it retrospectivley (?)

If AMD is faster than Intel or Windows XP is faster than Linux or Mac's then so be it, as long as that is what is actually happening since it shows the true work done, not what people are assuming is true.
(i.e. did anyone actually test to see if Intel CPU'd where not actually that much slower than AMD's for the same task/work unit at Rosetta OR that the linux system where actually slower at crunching here at Rosetta ? Just because it's the same CPU doesn't mean the code is just as fast under a different OS.

Other flaws in benchmarking are that the current 4/5day re-bench that BOINC uses
does not take into account when computers drop into a slower speed either due to going on to Batteries(e.g. laptops). The computer system starts to heat up and so throttles back (something seen with of the peg computers with inadequate cooling and again laptops) due to the heat that 100% CPU time creates. While CPU time is reported the same, actual 'work done' is slower.


Hope that give the developers of the new system some things to think about.

---

Team mauisun.org

Trog Dog so a try against the Barbarians or the All Blacks is the same as a try against the White Horse Pub ?

Fact : A new accreditation method will be implemented very shortly.

There is no need to carry on and on over optimised clients. No one is winning and the forum/project is losing.

If you play by the same rules it is :)

---