Akosf, I'm the one who started all this about Crunch3r's app. I did so because he kept a version (5.11 available for download, and it was being used by some) of a Beta app available for main seti use which had a multiplier of 7 instead of 3.35 (the current seti multiplier). Instead of removing the app and correcting 5.12, he took his ball and went home. There's more to the story, but that's the heart of the matter.

I don't understand this 'mutiplier-thing'. It seems to be imperfect. I usually work for industrial customers, they always want crazy-proof solutions. That means everybody have even chance. Somebody left a backdoor to Crunch3r and he took advantege of it.
Is this the problem?

I can assure you, that you'll not hear a peep out of me if you make a new optimized seti enhanced app, that has the right multipliers in it. Eric states he'll be changing the multiplier in a version for Beta very shortly. You'll have to decide if it's worth your effort to make one now and then change it shortly.

So, the multiplier... Sorry, but i don't want to adjust it.
Hm... What does this multiplier? ;-))

Basically LOAD_STORE_ADJUSTMENT scales the raw estimate of floating point operations performed in FLOP_counter before reporting to the BOINC client. All setiathome_enhanced apps are expected to use the built-in estimates so they'll get the same counts, the Scheduler simply scales the reported boinc_fpops_cumulative to make a credit claim. So the LOAD_STORE_ADJUSTMENT value in seti.h is intended to make the credits here compatible with other projects, although it's documented as "Adjustement to flops counter to make FLOPS include floating point loads and stores." The value in the 5.15 app is 3.35 and the adjusted value of 3.51 was checked in June 6, so an app built from current source will have the new multiplier.

                                                    Joe

I think it's a weakness of the third-party optimized/downloaded apps and clients.

[part off-topic?]
The utility BoincStudio (BS) of DocMaboul, still in alpha version, is mainly intended to the "farm-users". (It includes an administration console under Windows, and a new boinc manager which incorporates the new managment functions. Site here)
It includes basic security fonction. BS keep in a text list the MD5 hashs off all known boinc clients and projects apps. When installing or upgrading a new client or app via the console, if the computed MD5 of the exe is not in the list, it is added to by default, but a warning is displayed. In charge to the admin to verify that the file is that awaited, and inoffensive.
[/part off-topic?]

---

So, the multiplier... Sorry, but i don't want to adjust it.
Hm... What does this multiplier? ;-))

I believe the multiplier is "LOAD_STORE_ADJUSTMENT", Joe Segur might be able to give your more detail. I see it in CVS log for seti_boinc/client/seti.h
The adjustment to the "Load store" seems to follow what we call a multiplier. If I'm not mistaken is "load store adj X fpops cumulative" to derive claimed credit, and it's in the Seti Enhanced App.

See This Post by Joe Segur.

hope this helps

tony

I personally hope you do get involved with making one for us windows users.

I just examine some executables to learn new things.
eg. I found that the current FFT library routines are optimized only algorithm-level, not in instruction level.

If you came up with a faster FFTW3F, that would be quite something :o)
The fftw3f that comes with the standard app does not make use of SSEx, though...
It will if compiled with different flags.

Writing faster FFTW3F isn't mine goal, but BOINC downloaded 'libfftw3f-3-1-1a.dll' that consists SSE compiled routines.

Right, fftw has its own sse-autodetect code. That's why it contains sse code, but not sse2. Of course it runs on non-sse processors.

regards,

Tetsuji

Writing faster FFTW3F isn't mine goal, but BOINC downloaded 'libfftw3f-3-1-1a.dll' that consists SSE compiled routines.

And you could speed it up already? That's impressive.

Is there any chance of you looking at linux binaries, too :o)

I could mail you something from time to time.

Regards Hans

---

I personally hope you do get involved with making one for us windows users.

I just examine some executables to learn new things.
eg. I found that the current FFT library routines are optimized only algorithm-level, not in instruction level.

If you came up with a faster FFTW3F, that would be quite something :o)
The fftw3f that comes with the standard app does not make use of SSEx, though...
It will if compiled with different flags.

Writing faster FFTW3F isn't mine goal, but BOINC downloaded 'libfftw3f-3-1-1a.dll' that consists SSE compiled routines.

double post, sorry.

---

Akosf, I'm the one who started all this about Crunch3r's app. I did so because he kept a version (5.11 available for download, and it was being used by some) of a Beta app available for main seti use which had a multiplier of 7 instead of 3.35 (the current seti multiplier). Instead of removing the app and correcting 5.12, he took his ball and went home. There's more to the story, but that's the heart of the matter.

I don't understand this 'mutiplier-thing'. It seems to be imperfect. I usually work for industrial customers, they always want crazy-proof solutions. That means everybody have even chance. Somebody left a backdoor to Crunch3r and he took advantege of it.
Is this the problem?

Rather not. The multiplier was changed the exact moment when the new app went from beta to production.

Crunch3r released an old version, and refused to take it down after he was asked in a not so kind manner.
Things got worse after that and Crunch3r left.

I can assure you, that you'll not hear a peep out of me if you make a new optimized seti enhanced app, that has the right multipliers in it. Eric states he'll be changing the multiplier in a version for Beta very shortly. You'll have to decide if it's worth your effort to make one now and then change it shortly.

So, the multiplier... Sorry, but i don't want to adjust it.
Hm... What does this multiplier? ;-))

Nothing important. It just multiplies the FLOP-Counter that the seti app provides with a constant to determine claimed credit.

I personally hope you do get involved with making one for us windows users.

I just examine some executables to learn new things.
eg. I found that the current FFT library routines are optimized only algorithm-level, not in instruction level.

If you came up with a faster FFTW3F, that would be quite something :o)
The fftw3f that comes with the standard app does not make use of SSEx, though...
It will if compiled with different flags.

PS. virus infected app????? Please do tell.

It was only a question.
I think it's a weakness of the third-party optimized/downloaded apps and clients.

Regards Hans

EDIT: fixed those darn "quote" tags again. How about a WYSIWYG Editor :o)

---

Akosf, I'm the one who started all this about Crunch3r's app. I did so because he kept a version (5.11 available for download, and it was being used by some) of a Beta app available for main seti use which had a multiplier of 7 instead of 3.35 (the current seti multiplier). Instead of removing the app and correcting 5.12, he took his ball and went home. There's more to the story, but that's the heart of the matter.

I don't understand this 'mutiplier-thing'. It seems to be imperfect. I usually work for industrial customers, they always want crazy-proof solutions. That means everybody have even chance. Somebody left a backdoor to Crunch3r and he took advantege of it.
Is this the problem?

I can assure you, that you'll not hear a peep out of me if you make a new optimized seti enhanced app, that has the right multipliers in it. Eric states he'll be changing the multiplier in a version for Beta very shortly. You'll have to decide if it's worth your effort to make one now and then change it shortly.

So, the multiplier... Sorry, but i don't want to adjust it.
Hm... What does this multiplier? ;-))

I personally hope you do get involved with making one for us windows users.

I just examine some executables to learn new things.
eg. I found that the current FFT library routines are optimized only algorithm-level, not in instruction level.

PS. virus infected app????? Please do tell.

It was only a question.
I think it's a weakness of the third-party optimized/downloaded apps and clients.

I see here is a 'warzone' because of optimized apps.
Stupid rules and low organization...

What rules or organization would you propose?

---

I would like to examine the possibility only of the machine instruction level optimizations, it doesn't give more improvement than 15-20%.

I see here is a 'warzone' because of optimized apps.
Stupid rules and low organization...

Did anybody hear about virus infected apps?

I think I recall that there were/are some trojan versions of BOINC/SETI running around at one point.

There is this thread.

And this thread about a false positive.

And then somethimes a virus will pop up somewhere between the keyboard and the floor. ;-)

There certainly have been "viral SETI installers" but I haven't actually heard of a virus-infected BOINC app. -- doesn't mean it didn't happen.

One advantage of a "DIY" optimization kit -- you get the compiler from a known source, the app. source files from a known source, etc.

---

I would like to examine the possibility only of the machine instruction level optimizations, it doesn't give more improvement than 15-20%.

I see here is a 'warzone' because of optimized apps.
Stupid rules and low organization...

Did anybody hear about virus infected apps?

I think I recall that there were/are some trojan versions of BOINC/SETI running around at one point.

There is this thread.

And this thread about a false positive.

And then somethimes a virus will pop up somewhere between the keyboard and the floor. ;-)

---

I see here is a 'warzone' because of optimized apps.
Stupid rules and low organization...

Did anybody hear about virus infected apps?

Akosf, I'm the one who started all this about Crunch3r's app. I did so because he kept a version (5.11 available for download, and it was being used by some) of a Beta app available for main seti use which had a multiplier of 7 instead of 3.35 (the current seti multiplier). Instead of removing the app and correcting 5.12, he took his ball and went home. There's more to the story, but that's the heart of the matter.

I can assure you, that you'll not hear a peep out of me if you make a new optimized seti enhanced app, that has the right multipliers in it. Eric states he'll be changing the multiplier in a version for Beta very shortly. You'll have to decide if it's worth your effort to make one now and then change it shortly.

I personally hope you do get involved with making one for us windows users.

tony

PS. virus infected app????? Please do tell.

I would like to examine the possibility only of the machine instruction level optimizations, it doesn't give more improvement than 15-20%.

I see here is a 'warzone' because of optimized apps.
Stupid rules and low organization...

Did anybody hear about virus infected apps?

Yeah, often the Seti forum is hot.
But these last days, The Top... :(

Luckily, the SETI@home-AstroPulse Beta site is a haven of peace.

Did anybody hear about virus infected apps?

no, why do you ask?

---

I would like to examine the possibility only of the machine instruction level optimizations, it doesn't give more improvement than 15-20%.

I see here is a 'warzone' because of optimized apps.
Stupid rules and low organization...

Did anybody hear about virus infected apps?

10% faster compared to what ?

the official release of SETI or Crunch3rs optimezed Enhanced version ?

Crunchers version gave about 65% faster times, from around 10 hours per WU to around 4 hours per WU.
This 10% faster than official release is well, sorry but it's pathetic compared


And since I'm in it for the cause I don't care about this 5% more credit.
As it is now with the Enhanced version despite that I still use the optimized Crunch3rs version I get 50% less credit than earlier, so even this is pathetic.

What I do care about is crunching as many WUs as possible in the fastest time possible to find "the signal".

What I'm also surprised about that the developers at Berkeley is making this an issue.
They if anybody should be more concerned about the science and the speed of which WUs are returned instead of involving in petty things like forbidding optimized clients and debating credits.

What do you mean by 'Berkeley is making this an issue"?

As I see it Eric (others) are simply trying to set the record straight and clear up much of the nonsense being written. If you're referring to Berkeleys' opinion on optimized applications, Erics' statement "WE OFFICIALLY SUPPORT THE DEVELOPMENT AND DISTRIBUTION OF OPTIMIZED SETI@HOME APPLICATIONS" is IMO totally unambiguous. Similarly, his explanation of the choice of open source release under the GPL is clear, as are the rights and obligations of both distributors and recipients.

It was their idea of having this for the first absolutely stupid system of credits. The 'one credit for one WU' was an excellent system, this CS system is silly at best.
To have a consistency check for the WUs would have been more than sufficient for ensuring that nobody would be cheating.

[/quote]

Well this has been discussed to the death already. Yes 1 WU=1 Credit did seem to work for Classic - until you realise the scale of cheating that went on. It was a trivial matter to cherry-pick the faster units (low ARs) and not process the slower (high AR) ones, without penalty or even detection. Some form of credit for time taken had to be bought in.
Secondly, SETI isn't alone, there are many BOINC projects. How would you feel if the 10,000 cycle LHC units (<10 minutes) got 1 credit, the same as a 4 hour SETI unit? A system running only LHC might get 240 credits per day, one running only SETI gets 6 credits a day.

Calling something "silly" is making a comparison between different ways of doing something. What are you comparing the current credit system with, to make that judgement?

I don't like the credit system, but I don't have a better idea. This was what the developers came up with and I havn't yet seen a proposal for anything remotely close to a better idea, so I have a choice - accept it or stop doing BOINC projects.

Other circumstances may one day stop me doing BOINC projects, but not the credit issue or the other cr*p going on here now.

---

10% faster compared to what?

the official release of SETI or Crunch3rs optimezed Enhanced version ?

I had only the official 5.15 app.
I'm just testing Crunch3rs architecture-specific optimized apps.

Crunchers version gave about 65% faster times, from around 10 hours per WU to around 4 hours per WU.
This 10% faster than official release is well, sorry but it's pathetic compared

I would like to examine the possibility only of the machine instruction level optimizations, it doesn't give more improvement than 15-20%.

Eric and/or Pappa...

Can we donate any Beta resources to helping akosf or whoever in developing and testing optimized apps? Seems like it might keep people busy and interested in Beta while we wait for Astropulse or new SETI Enhanced apps to test. We've got a wide range of platforms to offer and it should also help keep the Beta userbase intact and crunching.

Just a thought,

Dig

Which windows version is the fastest?
... and how much faster than official 5.15 on different CPUs?

Ooh...now things are getting interesting. :) I don't think there are any optimized Windows Enhanced applications in wide circulation besides the ones that Crunch3r used to offer. If you do look at Crunch3r's clients, though, I wouldn't use the numbers from the quick reference WU on marisan.nl to judge performance gains--the performance profile of that WU doesn't represent SETI performance in general, at least not from my experience optimizing for PPC.

Thanks. So, i will do a comparison between different versions (5.15, 5.12SSE, 5.12SSE2, ...).
I see 5.15 isn't optimized, i got a 10% faster code after 10 minutes work.
Hm...

I'm impressed. 10% improvement after 10 minutes of work is amazing. I can't wait to see what you can accomplish when you start to really look at it.

Great job!

I'm impressed also! I've been hoping someone would take over where Crunch2r stopped! Thank you akosf for seeing what you can do.

Fred

---

http://www.teamstarfire.org/