A holiday season race: TAS-members vs. non-TAS-members in F@H

[Insidious]

Originally posted by: petrusbroder
OK, F@H-gurus: Is it OK if we noobs ask a few questions before the race?

1. Is there a difference between the CLI-client and the graphical client in matters performance (I think it is, but how big?)?
2. Can you configure the the graphical client to crunch the bonus-awarding WUs?
3. What kind of WUs can a PIII, a Athlon Thunderbird @ 1 GHz, and a Celeron @ 1.4GHz crunch? (I think that a Athlon 64 3000? can take it all ...)
4. What kind of memory requirements are there for the different kind of WUs?

Thats all for today, but I think that I'll be back.

IK, here we go (there are more knowlegable folders... I hope they correct me where I'm wrong)

1. Yes, there is a small difference (<10% I think) because of resources used to create the graphics. Also, most of the threads where I read about system difficulties, it was a folder using the GUI and it wasn't playing nice with their graphic display adapter. Most folders seem to prefer the CLI version.

2. You can add parameters to the command line that you start the client with to make some choices. And when you configure the client some other options are offered. such as; You can agree to accept large work packets (>5MB) (If you have 521MB or better RAM, say YES. Also there is choice for the -advmethods tlag that indicates you are willing to accept (quasi) beta work units. This one lets you receive some of the bonus Work Units (but does not guarante it).

3. The 'Mac-Daddy' (QMD) bonus work units will ONLY be assigned to Intel platforms supporting SSE2. ( The reason I quit F@H :| ) The computers you mention can crunch the 'regular' work units fine. (albeit slowly). Just don't use the -advmethods flag and answer NO to accepting large work packets in the configuration. You can also make a configuration choice to fold "timeless workunits" which are WUs with no deadline to completion.

4. The Bonus work units use more memory than the 'regular' ones. I think you can expect up to ~100MB for most bonus work units and twice that or so for the QMD (I don't remember how much they want but it is a lot) This is why the bonuses are applied. For the 'regular work units, expect to be using ~30MB

Hope this helps

-Sid

NOTE: F@H works great when installed as a service. To monitor the progress of these installatioins (running invisibly) there is a very nice utility called fahlogstats

Also, you can't choose what WU you will actually be assigned. You can only choose to exclude some (by not setting the flags). The rest is (or so they claim) random chance of assignment.

 

[GLeeM]

1. If the GUI client is left up open displaying, it will rob some CPU cycles. If it is minimized the difference is not really measurable. The core does the crunching. After the core has nine hours crunching, my client has two seconds CPU usage! It only comes into play at the beginning and end of the WU.

2. I haven't run the newest GUI so I can't help much here other than to make sure you know that Insidious had a typo - you need 512 MB to get QMD bonus WUs. Other bonus WUs need less.

With the newest CLI console client - 5.04 - you can choose how much ram you want the client to report to the Assignment Server so if some bonus WUs use more ram than you want it is easy to set a lower amount of ram to report and they will send appropriate WUs.

3. The computers you mention can crunch some bonus WUs, if you can get them. WUs are assigned on a first come basis, if you have the client set to recieve them AND there are any available you will get one.

It seems the Assignment Server (AS) spends a different set amount of time assigning WUs from each server (depending on science priority) and just keeps going around to each one in order. If a bonus WU server has recieved enough results back since the previous AS cycle then it will issue some WUs.

If I can get them on my PIII 1Ghz laptop I get 100 ppd from the 600 pointers and 140 ppd from those numbered 147* or 148*. Regular WUs get 50-70 ppd.

Any questions, please ask!

 

[petrusbroder]

Thanks for the answers. I have switched my two slowest computers to F@H. Now: can I somehow influence the servers so that I can get 147* or 148* WUs? :Q

 

[Insidious]

Originally posted by: petrusbroder
Thanks for the answers. I have switched my two slowest computers to F@H. Now: can I somehow influence the servers so that I can get 147* or 148* WUs? :Q

The only way I was ever able to "screen" WUs was to delete the ones I didn't want (there is a -delete xx flag where xx = designation of current work unit as given in the fahlog.txt file. (ie: LOTS of babysitting)

[rant]

(they know no one would crunch their 'normally credited' WUs if we could choose. I'm getting all worked up again just thinking about this skewed, manipulated, unfair, irrational, self-serving, joke of a scoring system F@H uses)

[/rant]

edit: Oh yeah, your question... If you use the -advmethods flag and answer YES to the configuration question about accepting large work packets (>5MB) you will have the best chance that is available to get the 14xx series WUs (There are a couple other series that also give well bonus'ed rewards)

 

[petrusbroder]

... which essentially means that this is only for my most powerful crunchers ... (P4:s, A64 etc.) OK. I think get it.
And I think I understand your "rant", Sid.

Thanks for your help!

 

[Insidious]

Yeah, the INTEL SSE2 capable platforms are the ONLY ones that will be truly competitive (if they are crunching QMD Work Units)

To the tune of: It takes 4 AXP machines (folding non bonus'ed WUs) to equal the daily point production of a single P4 crunching a QMD

(But the A64 will knock them off pretty well too, so don't give up on it totally)

-Sid

(I apologize for my rants.... I am still working through some... anger issues.... I'll do better )

 

[GLeeM]

Originally posted by: Insidious
[rant]

(they know no one would crunch their 'normally credited' WUs if we could choose. I'm getting all worked up again just thinking about this skewed, manipulated, unfair, irrational, self-serving, joke of a scoring system F@H uses)

[/rant]

They know no one would crunch their 'new technology WUs', (the ones that they are so excited about because of the extra science that gets done), if there was not some bonus for the extra resources that are used. Everyone would just delete them.

The researchers' aim is to find cures for diseases. They created this game we enjoy to get more people to offer their computers to help them. They don't understand about TeAm synergy and how we feel. But they do understand their science and truly want to find the next big discovery!

If they had gone the other route with 'new technology WUs' and asked people to volunteer more resources for the same amount of points, how many do you think would help them out?

As for the QMDs, it's too bad that Intel and AMD can't get along for the benifit of users.

The QMDs are benchmarked like all other WUs, even the other bonus WUs, and then the bonus is added just like the other bonus WUs.

They couldn't really start a new DC project just for QMDs, at least not until they have been tested.

Maybe now that they are tested they could start a new BOINC project just for QMDs!

 

[Insidious]

The pretence of "benchmarking" is debunked as soon as we begin talking about adding bonuses to some and ONLY SOME of their work units.

It is an absolute joke to pretend to be using a standard while at the same time arbitrarily adding more credit to work units when this "standard" process doesn't yield a convenient result.

If they are going to modify "benchmark" results each and every time it is convenient, then it is stupid to benchmark them in the first place! Well, unless they need some tool to help convince the gullible that this is a scientific process full of objectivity and fairness.

In a second observation, it is my opinion that the F@H developers have failed miserably in terms of distributed computing because they aren't able to (or didn't have the patience to try to) break up this large job into small enough pieces... Small jobs cabable of being processed on a computer without disrupting the normal operation of that machine. If a work packet grabs so many resources from my computer that I can't go on about my business then it is a failure. (This is to speak to the notion that bonus points are necessary to get people to participate because of excessive resource usage)

So this is why I believe many folders are being duped.

We are told that the developers must benchmark on a crippled (SSE2 disabled) machine because it's only fair because earlier work units were done way.... er, we didn't like the result on some units so we ignored it and made up our own credit value. Anybody see a contradiction here?

Not to mention the fact that the only plaftorm specific (Intel only) work unit just happens to have the largest bonus of any in the project... Does any entity (that just happens to have a reputation for donations of equipment, etc to college research departments) stand to gain if crunchers must buy their product to participte?

I am sorry that this argument is occuring here in the AT forums, but I just can't sit by and stay quiet when I believe so strongly that the F@H project is inherently flawed and (the part I can't forgive) the flaws seem to be self-serving.

 

[BlackMountainCow]

I did some 200 WUs for F@H before I quit. I absolutely agree with Sid on this one. Back when I quit, the big WU were giving good points but they also made some of my systems really sluggy because of the memory they consumed. Indeed, the idea of DC is to use IDLE cycles, not ALL there are.

Very well put Sid!

 

[Rattledagger]

Originally posted by: GLeeM
Maybe now that they are tested they could start a new BOINC project just for QMDs!

Folding@home releasing QMDs under BOINC? :Q
Don't hold your breath...

 

[xbassman]

Wow...there is sure a lot of anger here...... :shocked:

With the newest CLI console, you can configure both memory and CPU usage. That way it won't take over your system.

The last 3 DC profects I have participated in used a slow P4 for benchmarking. (gotta have some frame of reference.)

 

[Insidious]

Originally posted by: xbassman
Wow...there is sure a lot of anger here...... :shocked:

With the newest CLI console, you can configure both memory and CPU usage. That way it won't take over your system.

The last 3 DC profects I have participated in used a slow P4 for benchmarking. (gotta have some frame of reference.)

Did those 3 DC projects ignore the benchmark result for selected work units in preference for a credit value they found more convenient? (ie: ultimately assign a different point value than that dictated by the benchmark test.)

Also, did those 3 DC projects select, based on CPU manufacturer, who should get the Work Units which provided greatly inclresed point values?

These two points are the ones I believe are unique to F@H.

But we digress.....

Let the hijackers desist (myself included) and let this thread return to it's intended purpose....

A FUN race... I will arrive with a smile on my face and proudly fold with our TA F@H TeAm come race time.

-Sid

 

[petrusbroder]

Thanks Sid, for you opinions - which throw some lights on the project and which caused - at least me - to think about bonuses, credits and such. And thinking is good. It will make me decide in the future what to look for and which projects to crunch for.
Thanks also for your decision to crunch in this race, which I hope - will be more fun than the usual crunching. And it will be a holiday-race, a Christmas season race.

One wise man (Kenji Ota) once said: Love your enemies - it will drive them crazy! I certaily believe that - and I have seen it work!

 

[Insidious]

Thanks Peter

I may have a surprise (of the best kind) to bring to the race with me. Ya see, there is this big brown truck heading my way with 2 circuit boards on it.... unfortunately, $$ are going to dictate how soon I can find the right sized IC to fit in those empty sockets... If I remember correctly they have nine hundred and some odd pins on them.

:beer: MuHuhahaha :beer:

-Sid

 

[mrwizer]

Perhaps 939?

 

[Insidious]

:thumbsup:

 

[MDE]

Originally posted by: Insidious
Thanks Peter

I may have a surprise (of the best kind) to bring to the race with me. Ya see, there is this big brown truck heading my way with 2 circuit boards on it.... unfortunately, $$ are going to dictate how soon I can find the right sized IC to fit in those empty sockets... If I remember correctly they have nine hundred and some odd pins on them.

:beer: MuHuhahaha :beer:

-Sid

If you'd like an Athlon 64 3000+ for one of those boards, shoot me a PM. I've got a spare one lying around that needs a home.