Automatic Benchmarking

Heffed
Heffed
Joined: 18 Jan 05
Posts: 257
Credit: 12368
RAC: 0

RE: Heffed, I posted asking

Message 16117 in response to message 16116

Quote:
Heffed, I posted asking for information, and you are unable to give me any apropos to my question.


ap·ro·pos Pronunciation (pr-p)
adj.
Being at once opportune and to the point. See Synonyms at relevant.
adv.
1. At an appropriate time; opportunely.
2. By the way; incidentally: Apropos, where were you yesterday?
prep.
With regard to; concerning:
Apropos our date for lunch, I can't go.

I was making comments 'with regard to; concerning:' your question. ;)

Quote:
Your point is valid, your righteous indignation superfluous, your sarcasm noted. I see you care not to address the secondary issue of the tilted playing field, the benchmarking imbalance, an artificial foundation for credit, without which my artifice would not be motivated. Crunch well, my brother, and Thank you.

HeHe... Calm down.

Yes, I'm bent for sarcasm. Righteous indignation? Hardly... Merely playing devils advocate. I honestly couldn't care less about credits, therefore whatever you want to do in an attempt to garner more of them matters not to me. I'm merely pointing out that what you are doing can be viewed as questionable by the community at large, and it isn't really doing much in the grand scheme of things. In my opinion, not worth the effort in other words... Gather from that what you wish.

Tilted playing field, and benchmarking imbalance? Absolutely! People have been screaming about this since BOINC was in beta. It's light years better now than it used to be. The BOINC dev team is currently looking at the best way to base credit on the actual FLOPS reported by the science app instead of using benchmarks to define requested credit. :)

I was serious in my last post. You yourself mentioned that the optimized clients were requesting more credit than you. Why don't you just run one of them? You will request more credit than you are currently by doing your overclock dance, and it would be considerably less hassle. If you don't wish to pursue this avenue, you don't have to. I just thought I'd put it out there. Then my sarcastic wit came out and I added the bit about overclocking with the optimized CC. No malice intended.

SerVal
SerVal
Joined: 29 May 05
Posts: 3
Credit: 92736082
RAC: 0

2 Michael Roycraft: "Cut the

2 Michael Roycraft:
"Cut the baloney and give us ALL equal credit for each Einstein WU"
You eare about right !

Credit for WU = Const/(t2-t1)
where (t2-t1) - server time between send WU & receive result.
The rest cut no ice.
Benchmark results should not use in credit at all!

"you are unable to give me any apropos to my question"
Here is primitive solution:

1.Install Visual Studio Net.
2.download & unpack boinc_public-cvs-2005-08-29.zip
3.download & install wxMSW-2.6.1-Setup.exe
4.click on f:\boinc_public\win_build\boinc.sln
5.In file f:\boinc_public\client\cs_benchmark.c
edit this:
#define BENCHMARK_PERIOD (SECONDS_PER_DAY*5)

6.Build solution.Stop Boinc Service.
7.Replase Boinc files & libraries.
8.Start Boinc Service.
9. Start BOINCmanager.
10. Run benchmark.
(A lot of "optimization" also available, including return ANY benchmark result....)
If it's suitable for you...

Cupojoe
Cupojoe
Joined: 24 Feb 05
Posts: 23
Credit: 33068058
RAC: 0

Indeed, the amount of credit

Indeed, the amount of credit granted should be based solely on the amount of math (or work) contributed to the project (regardless of hardware or software platform, or whether someone is using an optimized client). The credit system for Einsein@Home is more convoluted than that used for SETI Classic (at least with SETI you could, for the most part, say "I did x work units at y TFLOPs per unit -- although I realize they doubled the amount of work to process a unit at some point). Einstein@Home should either make every work unit the same size (maybe within a couple percent either way), or, if work units must be different sizes, maybe create several classes of sizes that perhaps vary only a couple percent within each class, then tag that work unit with the size class to which it belongs. The current system is way too arbitrary: sometimes I get about 50 percent of the credit I usually get, even though I know my computer has done the same amount of math it almost always does.

5 and a half of 13
5 and a half of 13
Joined: 23 Feb 05
Posts: 36
Credit: 21194
RAC: 0

RE: Indeed, the amount of

Message 16120 in response to message 16119

Quote:

Indeed, the amount of credit granted should be based solely on the amount of math (or work) contributed to the project (regardless of hardware or software platform, or whether someone is using an optimized client). The credit system for Einsein@Home is more convoluted than that used for SETI Classic (at least with SETI you could, for the most part, say "I did x work units at y TFLOPs per unit -- although I realize they doubled the amount of work to process a unit at some point). Einstein@Home should either make every work unit the same size (maybe within a couple percent either way), or, if work units must be different sizes, maybe create several classes of sizes that perhaps vary only a couple percent within each class, then tag that work unit with the size class to which it belongs. The current system is way too arbitrary: sometimes I get about 50 percent of the credit I usually get, even though I know my computer has done the same amount of math it almost always does.


(Unneeded stuff removed)

There is hope at hand. Those who have been watching the BOINC news will know that a system to allow the science client to do it's own benchmarking has been proposed. According to the June 24 news the API calls have been added. Hopefully they will be in the next stable BOINC client.

PS Personally I find the credit allocation much fairer on Einstein, but that's probably because I get a higher amount that I claim ;)

PPS: I hope I'm not taking this too OT but does anyone know if Einstien plan too add thier own benchmarks to thier client if the API calls remain in the BOINC client?

Edit: Removed the stuff about the existing benchmark, which you guys already know better than me.

Need Help? Try the excellent Unofficial BOINC Wiki!
We are the BOINC. Prepare to be assimilated.
'anthrax beats WinXP' - The Register

Michael Roycraft
Michael Roycraft
Joined: 10 Mar 05
Posts: 846
Credit: 157718
RAC: 0

RE: RE: Heffed, I posted

Message 16121 in response to message 16117

Quote:
Quote:
Heffed, I posted asking for information, and you are unable to give me any apropos to my question.

ap·ro·pos Pronunciation (pr-p)
adj.
Being at once opportune and to the point. See Synonyms at relevant.
adv.
1. At an appropriate time; opportunely.
2. By the way; incidentally: Apropos, where were you yesterday?
prep.
With regard to; concerning:
Apropos our date for lunch, I can't go.

I was making comments 'with regard to; concerning:' your question. ;)

Quote:
Your point is valid, your righteous indignation superfluous, your sarcasm noted. I see you care not to address the secondary issue of the tilted playing field, the benchmarking imbalance, an artificial foundation for credit, without which my artifice would not be motivated. Crunch well, my brother, and Thank you.

HeHe... Calm down.

Yes, I'm bent for sarcasm. Righteous indignation? Hardly... Merely playing devils advocate. I honestly couldn't care less about credits, therefore whatever you want to do in an attempt to garner more of them matters not to me. I'm merely pointing out that what you are doing can be viewed as questionable by the community at large, and it isn't really doing much in the grand scheme of things. In my opinion, not worth the effort in other words... Gather from that what you wish.

Tilted playing field, and benchmarking imbalance? Absolutely! People have been screaming about this since BOINC was in beta. It's light years better now than it used to be. The BOINC dev team is currently looking at the best way to base credit on the actual FLOPS reported by the science app instead of using benchmarks to define requested credit. :)

I was serious in my last post. You yourself mentioned that the optimized clients were requesting more credit than you. Why don't you just run one of them? You will request more credit than you are currently by doing your overclock dance, and it would be considerably less hassle. If you don't wish to pursue this avenue, you don't have to. I just thought I'd put it out there. Then my sarcastic wit came out and I added the bit about overclocking with the optimized CC. No malice intended.

Heffed,

Sorry to take so long to respond. I've done a search of sorts and can't seem to find an optimized app for Windows running Einstein.

SETI is another story, there appear to be optimized apps for that, but I discontinued running SETI work 15 months ago, after turning in 3100+ WUs there. I really can't see the use of SETI, at least for contacting ETI. The distances and time factors against contacting an extraterrestrial intelligence, doing the handshaking to establish communication, etc reduce those chances to infinitessimal. Hopefully, we MAY find enough evidence to conclude that there is (or WAS, rather) some company in the universe, but given the extremely long time span involved in radio waves travelling immense distances at no faster than lightspeed and the likelihood that a civilization (including our own) will decay or be destroyed, I don't see it happening for several lifetimes, if at all.

I'm not trying to make converts to my pessimistic viewpoint here, I'm only explaining as best I can the rationale behind dropping out of SETI work.

microcraft
"The arc of history is long, but it bends toward justice" - MLK

Bill Sturgeon
Bill Sturgeon
Joined: 25 Aug 05
Posts: 13
Credit: 5370139
RAC: 74530

Are we after credits or

Are we after credits or joining in pushing the envelope of knowledge? Sorry I am just a construction worker that didn't go to college!!!

Michael Roycraft
Michael Roycraft
Joined: 10 Mar 05
Posts: 846
Credit: 157718
RAC: 0

RE: Are we after credits or

Message 16123 in response to message 16122

Quote:
Are we after credits or joining in pushing the envelope of knowledge? Sorry I am just a construction worker that didn't go to college!!!

Welcome to E@H! I believe that most of us are both, sir.

microcraft
"The arc of history is long, but it bends toward justice" - MLK

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5842
Credit: 109394736718
RAC: 35810496

RE: .... I've done a search

Message 16124 in response to message 16121

Quote:
.... I've done a search of sorts and can't seem to find an optimized app for Windows running Einstein ....

What Heffed was referring to are the optimised core clients on which you can run the EAH science app. It's the core client that does the benchmarking and not the science app. These optimised core clients were made for running the optimised Seti app but there is nothing to stop you running EAH and not Seti. The better benchmarks of the optimised CC will enable your completed results, still taking the same cpu time, to claim higher credit. Your current claims of around 70 would probably rise to around 75-76 or so. It is, however, debateable as to whether you actually get much benefit.

I notice you are running V4.37 of the core client. That's not a particularly recent version anyway so if you don't mind running older versions and really do want to claim higher credit without doing your current overclock dance, why not revert to V4.19 which has a benchmarking bug that will really allow you to claim quite a lot higher than even an optimised CC? Your current 70 claim would be transformed to a claim in the mid 80s. Once again, the other members of your quorum will mostly drag you down but you'll get an occasional "win" :). The other problem, of course, is the relative primitiveness and lack of features of the older GUI, although that probably means that fewer cycles are stolen from the science app :). It's also not a long term solution as sooner or later the move to version 5 will kill it off.

Cheers,
Gary.

Michael Roycraft
Michael Roycraft
Joined: 10 Mar 05
Posts: 846
Credit: 157718
RAC: 0

RE: RE: .... I've done a

Message 16125 in response to message 16124

Quote:
Quote:
.... I've done a search of sorts and can't seem to find an optimized app for Windows running Einstein ....

What Heffed was referring to are the optimised core clients on which you can run the EAH science app. It's the core client that does the benchmarking and not the science app....why not revert to V4.19 which has a benchmarking bug that will really allow you to claim quite a lot higher than even an optimised CC? Your current 70 claim would be transformed to a claim in the mid 80s.

Thanks, Gary, for the information. I was running V4.19 until I updated a few months ago. I like the additional features the change brought, and didn't notice any significantly longer processing times - my machine has only been able to get into the high 18000 sec/WU range throughout my 5 1/2-month workspan on E@H, and then only if I am not "distracting" it with work, games, surfing, IM client, etc.

BTW, "I notice you are running V4.37 of the core client...", true, but how can you see what version is running? I'm curious. Somehow, the client has made a mistake in identifying my CPU characteristics. Although my CPU always shows up as "unidentified Athlon XP" because it is a notebook CPU "Athlon XP Mobile 2600+" running in a desktop, the CC reports it having 976.56KB cache, while it has only the 512KB of any AMD Barton core.

Thank you again for your time and attention to my question. I shall carefully consider the options that you have presented.

microcraft
"The arc of history is long, but it bends toward justice" - MLK

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5842
Credit: 109394736718
RAC: 35810496

RE: RE: I notice you are

Message 16126 in response to message 16125

Quote:

Quote:
I notice you are running V4.37 of the core client...

... true, but how can you see what version is running? I'm curious.

Try clicking on the resultID of one of the results in your results list and you will see that the version of BOINC is noted there.

Quote:

Somehow, the client has made a mistake in identifying my CPU characteristics ...

Please don't blame the poor core client as it is only passing on what the OS has told it and the OS is only passing on what the BIOS has told it and the BIOS (for a desktop MB) is probably rather confused about this Mobile AMD cpu that doesn't match anything it knows about :).

Quote:
Thank you again for your time and attention to my question. I shall carefully consider the options that you have presented.

You're welcome. Please realise that there is nothing new or earth shattering in any of the information I gave and it is all freely available by searching through previous messages using the search function. The most important thing is to have fun while contributing and to take pride in playing a small part in the advancement of our understanding of the universe.

Cheers,
Gary.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.