A question of manners

[dimilunatic]

Spending time day by day lurking around the forum, I see a rising tendency of disrespect to the user -in general, the person who is generously offering his/her hardware to help science get results. Heck, not science; Stanford University, but let's not go down that path.

It seems to me that most of the time when someone has a suggestion or needs help with something -apart from getting the clients to work-, the answer is "suck it up and live with it" more or less. Is this a proper way to trust a generous individual who potentially puts their hardware in danger? Not to play the preacher here, but I'd think that donors are generally treated with respect, especially these days when the power of the internet can knock out anything within a period of months.

First of all, many people around here have asked the Pande Team to program the servers as to distribute whichever WU is best for whichever family of cpus/gpus. Still, this is not even on a to-do list for the next years, all strangely. The fact that some particular WUs only give EUEs to my video card, when no others do so is a clear example of how my card has limited capabilities towards them. And yes, I have used every single 3rd party program to stress it and had no problems whatsoever. No driver problem, no overclocking at all. It's not only a matter of stability either. It's also a matter of speed. People who fold on PIIIs and people who fold on Intel Core 2 Duo's should not be given WUs of the same size, or the old-tech guys will sooner or later bail out. In addition, spending $700 for a new processor -as many rush to suggest- isn't so much a possibility for everyone these days. *economic crisis hinthint*

Secondly, I've been folding for a bit more than a year and have noticed the problem of people who want to transfer their points to a new team, etc. Sure, if everyone starts doing that, the servers will probably crash. However, there is always an exception to every rule. What if the person who started a team is now gone from the face of the earth and the team info needs to change? "Duh, open a new team and start folding." Well, it's not easy to say that to someone who devoted their valuable time, working to see their team rise and suddenly has to start from scratch. Also, what if someone of the medical/biological/pharmaceutical career, who wants to put FAH in his CV, was n00b enough to start folding with a ridiculous pseudonym and now has a couple million points or so? I believe that these are some examples of sensitive situations where exceptions to the rules should happen.

These are my thoughts for now. I don't mean to blame anyone. I absolutely love folding@home for what it's taught me and what it offers to humanity. That's why I feel the need to help this project by submitting my own constructive criticism.

I'm waiting for your responses, whether positive or negative.

[patonb]

The problems I can easily see with giving people choice of what to fold is the people who would abuse it for point reasons... i can get upwards of 8600ppd out of the 353 ptrs. But then the secondary issue is wus are put out based on the science that needs done, so all the non popular units won't get done and delay work overall.

Point transfers shouldn't be allowed at all. As been stated, it doesn't take away from a users total to change team, and really only hurts the team that becomes inactive. Just because a team isn't active, doesn't mean that team contributions didn't matter, and to take those points away, takes that teams contribution effort away.
You need to look at the offical stats, as 3rd party stats dont show the whole picture. The offical stats show ALL your contributions associated to your user, and if you were dumb enogh to put a username you aren't proud enough to show in public, than thats your own fault. Correcting your childness shouldn't be Stanfords ploblem.

[7im]

With respect, what "to do list" are you looking at to make the determination that improving WU distribution to better match hardware is not on that list?

And no disrespect, but some number of EUEs are expected. And WU selection is the least common cause, while appearing to be the most common symptom. http://fahwiki.net/index.php/Early_Unit ... WUs_EUE.3F Without knowing the error types, your specific hardware, and the specific work units or project numbers, I couldn't begin to guess at the cause. It could be a WU based problem, or could be something else. Did you start a discussion here to get help for resolving the EUE issues, or seek recommendations for minimizing the problem?

As for the Pand Group policy for non-intervention on team and user points, it is 10 years old. Some would say it works well because it has lasted 10 years. Others would say it's archaic and needs to change. Yes, there is probably a better way to handle this. But Pande Group isn't going to hire a new person to do this, so an existing person will have to spend less time on the science of the project to do this work. Is the cost of that better way worth the expected benefit? How does one determine such a thing?

As we all know, implementation is much more difficult than inspiration.

[BuddhaChu]

If the Pande group allowed users to select whatever WU they wanted, everyone would pick the one that gives the most PPD on their hardware. That's human nature. To avoid that scenario, the client gives you a random WU from the pool of available WUs thereby giving each project the same amount of CPU resources.

Just because your particular video card can't work with a specific WU doesn't mean that entire class of card can't. If Pande has data that says if we give 1000 nvidia XXXX cards, WU YYYY and 99% are returned fine, something is wrong with those 10 cards that can't finish that project's WU. On the other hand if 1000 Nvidia ABCD cards are given WU XYZA and 60% are never returned and the forums light up complaining about WU XYZA, I'd say it's safe to say WU XYZA or the client has "issues", not not the Nvidia ABCD cards (this is why WUs get beta tested).

If somehow you and a significant majority (whatever that may be 20%, 30%, etc) of users of your particular card have an issue with a particular WU, I'd say you have a point and that WU should be pulled. If it's a small minority (you and your friend for example) I'd say it's your hardware that's at fault.

Confucius say, "Look inside for the source of your problem, not outside".

EDIT: I started typing this when there were no responses, but I type with two fingers and kick in the thumb occasionally for the spacebar. My 4th grade spelling ability doesn't help much.

[dimilunatic]

With respect, what "to do list" are you looking at to make the determination that improving WU distribution to better match hardware is not on that list?

Vijay Pande's blog, announcements of the forum, Vijay Pande's twitter etc etc.

And no disrespect, but some number of EUEs are expected. And WU selection is the least common cause, while appearing to be the most common symptom. http://fahwiki.net/index.php/Early_Unit ... WUs_EUE.3F Without knowing the error types, your specific hardware, and the specific work units or project numbers, I couldn't begin to guess at the cause. It could be a WU based problem, or could be something else. Did you start a discussion here to get help for resolving the EUE issues, or seek recommendations for minimizing the problem?

Yes, I did start a discussion, which hit 2-3 pages. No direct answer had been given, other than that one specific WU would always crash when received. Period. I remember posting fahlog on fahlog on that thread and nobody had a clue what to do. Also, after some search on this forum, I found more people with the exact same problem on the exact same WU, who still had no clue as to how to fix it.

If the Pande group allowed users to select whatever WU they wanted, everyone would pick the one that gives the most PPD on their hardware. That's human nature. To avoid that scenario, the client gives you a random WU from the pool of available WUs thereby giving each project the same amount of CPU resources.

I never spoke about cherry-picking. All I asked was that the servers themselves decide which WUs are best for which family of Cpus/gpus etc.

[BuddhaChu]

I never spoke about cherry-picking. All I asked was that the servers themselves decide which WUs are best for which family of Cpus/gpus etc.

Your initial post mentions the clients doing the selection, not the Stanford servers:

First of all, many people around here have asked the Pande Team to program the clients as to receive whichever WU is best for them.

[dimilunatic]

I never spoke about cherry-picking. All I asked was that the servers themselves decide which WUs are best for which family of Cpus/gpus etc.

Your initial post mentions the clients doing the selection, not the Stanford servers:

First of all, many people around here have asked the Pande Team to program the clients as to receive whichever WU is best for them.

Sorry for the misunderstanding. I'll go and rephrase that.

[7im]

The lack of discussion for a specific feature on Vijay's blog and twitter accounts does not equate to a lack of development for that feature. Correlation is not causation. Just a guess, but when Pande Group hires a programming contractor to rewrite the fah client and server code from the ground up, that would seem to me to be the best time to implement all features already requested (if feasible). And better WU to hardware matching has been a perpetual request. But please consider that hardware changes faster than the client software, so that request is always going to be a moving target. Please also consider that if the client gets too specific about WU selection, then some work units may never get processed. PG follows the science, not the buying preferences of the contributors.

I will try to find your thread. If the problem was with one specific work unit, then according to that WIKI article, it's not a problem with your hardware. It's just a bad work unit. If the PRCG numbers were posted in your thread, we can have a Moderator mark that work unit as bad so it never gets sent out again (if that hasn't been done already).

[dimilunatic]

I will try to find your thread. If the problem was with one specific work unit, then according to that WIKI article, it's not a problem with your hardware. It's just a bad work unit. If the PRCG numbers were posted in your thread, we can have a Moderator mark that work unit as bad so it never gets sent out again (if that hasn't been done already).

It's ok. That was an old Wu many months ago and I haven't gotten one like that for some months, so don't worry. I only said that as an example, since there might be more of them out there.

[P5-133XL]

My first reaction is that if you have a problem, that you want fixed, with the behavior with a specific person, you should be contacting that person; or that persons superior (i.e. a moderator or the sites owner), privately, depending upon the specific behavior and how it affects others. At no point is it ever appropriate or even useful to use the court of public opinion to attempt to change a specific persons behavior (at the level of manners).

The next observation is that you attribute a rising tendency of disrespect to the site and attribute it all on one specific individual. If the problem is the behavior of the site or the organization that is one thing, but if the problem is a specific individual then that is a totally different thing. Well which is it? If it is one individual, then if you want it fixed, then you have to name the individual if you want that individual's problems fixed. You can't keep it in the abstract because it can't be fixed if no one knows who. A person's ability to rationalize his own behavior is infinite. If you don't name names, then it will always be someone else from the perspective of the offender.



Now lets get more into the specifics of your issue. You have a problems with a card that EUE's under specific conditions. You have repeatedly tried to get help and it has been unsuccessful. Sorry, but not all problems are fixable by us. We can't fix broken HW and we can't change code. that is not this sites purpose! We are not attached to Stanford. We don't control the assignment servers. We don't write the software. We don't build video cards. We don't even make policy. This site is just a group of volunteers trying to offer verbal support for folding through these forums. We offer setup and configuration advise only. At what point should a problem be recognized as impossible to fix by us? The next question is how should that impossibility be addressed other than we can't fix your specific problem so you will just have to live with it or not fold?

You are welcome to add a suggestion for future improvements to the code and I'm sure that Stanford will take you suggestion into consideration in developing future code. People that do write code, do frequent this place so it may be useful.

The second problem you have is the transfer of points to a new team. That is a matter of policy. That particular decision was decided a long time ago that It is not going to change. It is not that the servers will get overloaded. Rather there are three issues with it. The first is that it has to be done by hand and the people that deal with budgets are not willing to pay for that feature. The second is that the ability to transfer points from one team to another will create competition, among teams, for specific people that are very high point producers to move the rank of the team. The ability to allow these people to move their points from one team to another produces totally invalidates team rankings if you can produce points for one team and them move them to a different team, on a whim Third, I do not believe that Stanford wishes to create an marketplace where specific individuals can extort their team to stay with them, collect something to move to another team, or for a team to bribe a person away from one team to their team.

P.S. I too started typing my response before there were any other responses...

[Pette Broad]

If the Pande group allowed users to select whatever WU they wanted, everyone would pick the one that gives the most PPD on their hardware.

Well, that may be the case for many, but I prefer smaller units regardless of the PPD. I'd much rather do a 12 hour unit for 120 PPD than an 80 hour one at 300 PPD. Several members of my team have also expressed the same opinion. The problem is that you can only do what's available and I don't expect Pande Group to make specific units available.

The second problem you have is the transfer of points to a new team. That is a matter of policy. That particular decision was decided a long time ago that It is not going to change. It is not that the servers will get overloaded. Rather there are three issues with it. The first is that it has to be done by hand and the people that deal with budgets are not willing to pay for that feature. The second is that the ability to transfer points from one team to another will create competition, among teams, for specific people that are very high point producers to move the rank of the team. The ability to allow these people to move their points from one team to another produces totally invalidates team rankings if you can produce points for one team and them move them to a different team, on a whim Third, I do not believe that Stanford wishes to create an marketplace where specific individuals can extort their team to stay with them, collect something to move to another team, or for a team to bribe a person away from one team to their team.

Great summary

Pete

[BuddhaChu]

My example was the "sinister, low road approach" some participants could take if given the opportunity. Much like you, if I I was given a way to pick my own flavor of WUs, I'd chose ones that heat up my GPU less thereby keeping the fan slower and quieter and still give me a decent PPD. My 8800GT gets LOUD on some WUs and just purrs nicely on others.

[Wrish]

On the subject of to-do lists at PG:

PG is a small team, not a programming powerhouse. There are a dozen of them, and a million of us (not quite, but you get the point). Most of the feature suggestions I've heard take lots of R&D and programming time on their end. They only just recently hired a company to code for them so that what team they have can focus on the science and publish results. For a small team, they have big decisions to make. Is it better spending time to troubleshoot problem projects here and there, or to rewrite the whole client on a more stable/efficient platform?

Some of the requests take a special amount of effort for the expected benefit. E.g.: Some people have been EUE'ing on GPUs. Some of these EUE projects I've caught finishing on a closely related GPU. It's probably very technical to figure out why that person's GTX260 or GTS250 keeps failing while my GTX280 doesn't (why hasn't it been the other way around?). Can a client even distinguish the stepping versions of these cards? E.g.: Some people complain that a certain class of projects runs hot on some cards, while others work slower and cooler. Should we have PG query the user whenever a new project appears, or do we already have adequate tools like GPU_IDLE and utilities to tweak clock/voltage and monitor temps? They are called high-performance clients for a reason. Sometimes we have to babysit them and apply our technical knowledge, and in turn we're rewarded with massive points. You're probably very good at automation and configuration or very lucky to have a high perf client run unattended for a month straight like the uniprocessor clients often do.

Also, what if someone of the medical/biological/pharmaceutical career, who wants to put FAH in his CV, was n00b enough to start folding with a ridiculous pseudonym and now has a couple million points or so?

Then I'd say they're just as n00b for hoping a recruiter would count folding experience alongside actual medical experience. If they're IT, different story perhaps, but PG and all us folders could really use some coding expertise.

I know PG has made exceptions to their policies, but very few. They totally moved someone with half a billion points from one team to another. Most of their side work is combating fraud. As soon as they give one handout to a mere mortal, though, the whole flock of us swarms them with demands.

What the team system needs isn't flexibility to override ownership and hop teams with previous work. As they stand, teams are insecure; you don't need an invite to insert your name onto a team; conversely, there's little assurance that the team you're contributing to won't suddenly change its name w/o your permission/vote.

[shdbcamping]

The lack of discussion for a specific feature on Vijay's blog and twitter accounts does not equate to a lack of development for that feature. Correlation is not causation. Just a guess, but when Pande Group hires a programming contractor to rewrite the fah client and server code from the ground up, that would seem to me to be the best time to implement all features already requested (if feasible). And better WU to hardware matching has been a perpetual request. But please consider that hardware changes faster than the client software, so that request is always going to be a moving target. Please also consider that if the client gets too specific about WU selection, then some work units may never get processed. PG follows the science, not the buying preferences of the contributors.
I will try to find your thread. If the problem was with one specific work unit, then according to that WIKI article, it's not a problem with your hardware. It's just a bad work unit. If the PRCG numbers were posted in your thread, we can have a Moderator mark that work unit as bad so it never gets sent out again (if that hasn't been done already).

It does not need to be as difficult as that. Have the client submit an individual fingerprint to the AS. Have the AS check that fingerprint for WU's assigned and not turned in. If an available WU does not have an XX% return rate then server go to next available..,.. etc, etc. until a relative assuredness of a successful return of a WU can be met. It could still try the hierarchy on all clients.

Also, if a particular WU is not finding acceptable clients it will be a Flag for Pande Group to reevaluate the WU class.

Sounds like 'win-win' to me.

Sean

[v00d00]

My idea of error checking would be if a persons client scraps workunits from project x, then why keep giving out that project to them, simply mark that user id as being unstable and assign them something different, eventually it may be that the client in question cant process anything due to instability in which case there would be no work left for that client to do. But at the same time it wouldnt be tying up server resources by submitting scrapped workunits, so it would not be affecting the rest of us. At that point the only thing to be done is for the user to reconfigure their machine so it isnt unstable, and lets face it most of this comes down to badly maintained hardware, bad hardware or overclocking. For a person to start folding again on that hardware they would have to go to a web page on stanford and input there username and passkey to reset the system. But if a person repeatedly did this and it failed, then they would be locked out until a PG member reset there status, which brings me to point 2.

What i would like out of PG, is a test program, similar to stresscpu. One that could download a special workunit that only ran to maybe 10 frames but absolutely hammered your system, with available workunits for Uni (all cores), SMP (all cores) and GPU (all cores), and then gave a result, so you could test for instability before it became a problem. Something like that with ihaque's memory checker also built in, so basically a test suite for people who want to run folding. If it comes up ok at the end theres a good chance you can fold without an issue. If it doesnt well you attempt to fix the problem, or post on here asking for help.

Optimised work units for optimised cpu cores would be nice as well, but its a lot of work, and wouldnt you all rather have stats that work, work units that are available to download, clients that are stable.

In the end if you want to pay for Vijay to have a codemonkey to add new backend functionality, no one is stopping you. Make a donation, and im sure it can be sorted out. At present i think the system is fine for the most part, and only really want an SMP3 client that runs on 32bit linux clients.