Fall of a Titan

It's sad that Stanford is driving away people from it's own project. Hope they fix it soon.

 
Yeah I had been seeing him going that way over there for a little while. I thought his job was about to change to..........

But his comments about ram usage have been spot on.
 
Thats a bummer.

Does the memory issue that hes talking about only affect dually rigs? I havent really read much about it. I havent had any real problems yet other than old hardware crapping out on me.




 
Well the RAM requirements have been going up (look at QMD's) and RAM release is an issue for the client....so yeah it affects everyone....just more noticeable on non-dedicated boxes and on duallies.
 
It sounds like he may have been getting some beta work units that aren't out in the wild yet that use a bit more RAM than the normal ones most people are seeing. Either that or it's QMD cores he's having problems with. I don't know about the QMD cores since I don't have a machine that can take them. The P4 I have at work has 512 meg of RAM but it also has integrated video which uses I think 8 meg so the client only reports it as having 502 meg of RAM.

I don't completely blame him for having a problem with this. I would also like to see a client that has a memory usage setting just like the processor usage setting. I wouldn't think it would be all that hard. Just have some setting that the client would only report that the system has only so much RAM instead of what the system actually has up to what the system actually has.

 
The math:

1 QMD core can equal up to 325+ meg RAM usage (150-325+ meg depending on the QMD WU)

2x Dual Xeon server with 4 instances (HT) of large packet WU assigned= ~1.2 Gig possible of RAM used just for folding.

1.2 Gig of RAM is alot, however this is really not the issue. One QMD core alone, can saturate the memory bus of a top of the line server all by itself. Running 4 will crush the preformance of the server. Many folks run one QMD core and a normal/timeless WU on the other CPU's, I think this was what OC-X (OC-AMD) was doing. Still a single QMD can slow things down.

Basically, my understanding, is a client with the options of timeless/normal/large/QMD memory usage was requested. With large WU's = ~125 meg RAM used at most.

This is just my understanding of the issue. I've got this same problem and I'm forced to run normal WU's vs. large WU's because I just can not allow a QMD to fall on certian boxes.
 
I have spoken on this subject before. Stanford for some time now has and continues to exhibit a total disregard for where all there work gets done.

This project started out with the intent of using your spare CPU cycles to accomplish a goal. Many of us went way beyond that and bought, borrowed, borged many machines in hopes of finding “the cure”.

As time goes on it seems the brains behind the project have made huge assumptions about the computers we run. Yes, many of us built and continue to build faster and faster machines and whether it was for the points or sake of the project didn’t matter, we just did it and were happy to do it.

Now we are faced with work units that require real “Work Station Grade” machines. From where does Stanford think these machines will come? Certainly not the average home user. Stanford itself can’t provide any number of this type machine, what makes them think we will, or should?

Somewhere pure arrogance has set in at Stanford. I find it pretty typical of today’s colleges and schools (I’ll hold off on any liberal speeches.). They want and expect to get it all, but always at someone else’s expense.

Will I stop folding, no. I really believe in the project. Do I think Stanford needs a bit of an attitude adjustment, absolutely.

 
I've noticed while being in college that sometimes people that only do work in that college loose touch with what is really going on. So they think because the big gamers are runnig 2GB of ram that all our boxen are dedicated to F@H so they figure they might as well use it.

Shoot I'm running on 384MB because of my mobo crash and F@H is taking a good chunk of it. I hope they realize this and can fix it soon.

 
Hey all,

I'm not leaving the community, just not using my machines until Stanford can do a better job of using what is available to them without killing the performance of our hardware.

The issues isn't with QMD's alone, but QMD's are a big part of it if not worse. Right now the client can detect how much RAM a system has, but can't detect how many CPU's it has... so if I run 2 CPU's with large WU's... I can get a mix of WU's that will simply bury a box. (HT P4's will see this also)

About the time QMD's were released they mentioned how much RAM newer WU's would take... this was 6 months ago if not longer... I raised the flag that RAM, unlike CPU utilization doesn't back down if the system needs it... so we need a seperate flag for large WU's vs memory intensive WU's... 6 months later, what I said came true and I have beta units using 800MB's of ram each on a dual Xeon with 1GB of ram and not using HT. The box is useless.

Stanfords fix is to run only a single large WU on a dually until a new client is realeased, my idea of a fix is to have the client know what it has availble to use or let us, the end users set that so we are using the box toit's full capacity. It makes no sense to me to run a single large WU and a regular Wu on a dual 3.6Ghz Xeon when I can run 2 or three "normal" large WU's on the same box and get them back faster... I have stacks of new dual Xeon's just sitting there doing nothing because you can't get a good balance of WU's on them.

Right now ,80% of my machines are off the project... if I can work with Stanford to find some type of fix until a new client comes out, I will... but I won't keep band-aiding my machines for an issues thats been known for months.

Cheers!
 
Sound s to me like they need a unified client. What I mean is a client that can Dynamically Scale Cpu & Memmory on demand. It would be great to install it as a service and if you got 1, 2 or 32 CPU's in a machine, it folds like gangbusters if the machine isn't being used heavily, but back off the CPU's as they're needed for other things, and Cap the WU's memmory requirements...

But that's my .02c
 
I hope the client is sufficiently amended in its next iteration that you can come back full force and without concern that F@H's memory usage will cripple your systems, OC-AMD. I support your position on this entirely.

I began to compose a comment on the sorry state of affairs with regards to this issue in the 'official' forum, but in this case I think discretion and prudence are the greater part of integrity. The Pande group would do well to recall as much with respect to their position as beneficiaries of the generosity of the F@H participants.

 
Wow, just wow. This is never news you want to see first thing in the morning. Here's to hoping a happy compromise, as thats a ton of work that could be done, but isn't. :(

 
I had one of those beta units just the other day, in fact it completed one the other day giving me a few points.

Really didn't see a drain on the memory, but I don't overclock them either. I remember the Work Unit being on my dual P3 xeon, just before I switched out the processors with a bigger cache. It was on a pc with about 2x 500mhz 512kb cache, about 512 mb of pc100 ecc memory. I had this machine set a timeless workunits and just happen to get this beta.

But I'm sorry to hear you bow out of Folding.
 
Wow... though I understand, it's not something I like to hear (anyone pulling machines off of F@H/UD)... I hope that Stanford can find a way to solve this issue... In the end it's all for the betterment (sp?, word even??) of humanity...

How can we help?


Keep on Folding!! For the [H]orde!!

 
Wow, bad thing to hear, but I completely understand. I use Athlon XPs and P3s, so I don't get any QMDs, but I have still seen some WUs take up a monster chunk on some of my 512MB systems.

I wish they would also priortize WUs better to architechtures. Stop assigning my P3s tinkers that take forever to finish and send them the way of the AMDs, and give my P3s some more 600 pointers, since they crunch through them much better. My AMD machines sometimes like to pick up 62 pointers that take a freaking day at 2GHz, which I would guess go a lot faster on a P4.
 
I'll second optimizing WU to processor. Every time my P-M gets a tinker, my production drops to a quarter of what it would with a Gromacs WU (for that CPU). A 600 point gromac takes about 2-2.5 days to complete on a P-M and a tinker takes around 3 days. I juset cringe everytime I see them hit it.

On the memory issue I have to agree. Especially with dual cores showing up.

I am setting up a Dual Opteron next week and having to plan to put 2Gb of memory in the system in order to deal with this issue.
 
What a drag. I haven't personally experienced this (5 of the 6 machines are AMD, and the other is a Intel laptop), but I can see where OC-AMD is coming from. It surely would be nice to have a more configurable client. I don't know how much work that'd be, but I suspect it wouldn't be all that much. Certainly better than losing a crapload of computing power from disgruntled users.

Maybe we ought to complile a list of users and how many ghz they were donating, who are now stopping in protest. Send that list to Vijay with a note explaining what's going on and why. Couldn't hurt.

Or what if we, the team, turned off our machines for say, a week, in protest, and tell Vijay exactly why. They'd have to notice if the entire Horde, or nearly all, stopped producing. Even better if we got OCAU involved too.
 
If this starts to cripple my machines, I will stop folding.....Cant have people work on slow machines, it's anti productive....



 
yes, I've noticed that I have a lot of problems with F@H running on my A64 gamer while I try to play BF2. I built the machine for gaming, not folding, so if the problem continues, I will have to remove the F@H client. (I can't get any of the net start/stop commands to stop F@H so I can play lag free, BF2 is a huge memory hog, and with only 1 gig of ram (I can't believe I just typed that!!!) I will use the PC for what I built it for, and reduce my output considerably since that folds like no tomorrow. It's sad, but when you help someone out, it seems like it suddenly becomes your job!!! As it is now, I have to turn off EMIII just to play BF2 without crashes!!! Stanford, get your act together and remember that all of us are doing you (and hopefully all of humanity) a favor.
 
Ronbo said:
yes, I've noticed that I have a lot of problems with F@H running on my A64 gamer while I try to play BF2. I built the machine for gaming, not folding, so if the problem continues, I will have to remove the F@H client. (I can't get any of the net start/stop commands to stop F@H so I can play lag free, BF2 is a huge memory hog, and with only 1 gig of ram (I can't believe I just typed that!!!) I will use the PC for what I built it for, and reduce my output considerably since that folds like no tomorrow. It's sad, but when you help someone out, it seems like it suddenly becomes your job!!! As it is now, I have to turn off EMIII just to play BF2 without crashes!!! Stanford, get your act together and remember that all of us are doing you (and hopefully all of humanity) a favor.
Click to expand...

Are you using console or graphical version? I also play BF2 and have to shut down the graphical client when playing. I routinely do this with games nowadays...
 
Yeah when I play any games, I turn off F@H and any other progs that use up memory..
 
The graphical client and I don't play well together - it typically crashes any game if its one when I'm trying to play. It got bad enough on my gaming rig that I just took it off my computer. What really annoyed me is that I emailed the stanford team the runs F@H and didn't even get a stock email back.... :mad:
 
So, the position for #1 Folder is now mine for the taking!!!!

j/k the RAM usage is getting too counterproducxtive I will have to stop folding as well. Luckily most ppl are A) not here during the summer B) don't notice C) upgraded from a 450mhz machine to a 2.8ghz machine and don't notice

 
If I set everything to small packets, and avoid large packets and timeless, how likely am I to get one of this massive WUs? Anyone know?
 
DamienThorn said:
If I set everything to small packets, and avoid large packets and timeless, how likely am I to get one of this massive WUs? Anyone know?
Click to expand...
100% you won't get one.
 
marty9876 said:
100% you won't get one.
Click to expand...

K...that's what I was hoping. I can't afford to have the computers at work to slow down at all. The second that users get legitimately angry at F@H is the instant it comes off all my computers.
 
I'm new to folding, and new to the [H]orde. But I cannot run F@H when my kids play games on my game machines. I taught them how to quit and restart folding.

"Yeah when I play any games, I turn off F@H..."

The games just will not play with F@H running. And I'm not talking about modern expensive cutting edge games. My daughter cannot play Harry Potter with F@H running on a 2.8Ghz P-4 with 1GB of memory.

Since I am new to folding, I kinda don't know what is normal and what is not. But I could not imagine running this at work and having these problems taking place in a business environment. The computers at work are for work, not folding.

 
Pocatello said:
I'm new to folding, and new to the [H]orde. But I cannot run F@H when my kids play games on my game machines. I taught them how to quit and restart folding.

"Yeah when I play any games, I turn off F@H..."

The games just will not play with F@H running. And I'm not talking about modern expensive cutting edge games. My daughter cannot play Harry Potter with F@H running on a 2.8Ghz P-4 with 1GB of memory.

Since I am new to folding, I kinda don't know what is normal and what is not. But I could not imagine running this at work and having these problems taking place in a business environment. The computers at work are for work, not folding.

Click to expand...

Are you running graphical client (pretty pictures) or the console service? Use the service if your not, very few problems there. Check the FAQ's here for more info.
 
Ronbo said:
yes, I've noticed that I have a lot of problems with F@H running on my A64 gamer while I try to play BF2.
Click to expand...


I noticed that too..I have to shut down FAH when I play BF2...... But I can play BHD and a few other games while it is running.... I figured it was because BF2 was a memory hog...I have only 1 GIG :) Im upgrading to 2 this weekend, Ill see if that makes it posible to run FAH and BF2 at the same time



 
So with datacenter server there are options to limit memory in a process group.

MSFT Article

I did a quick google but couldn't find anything. Seems logical that there would be some app that can control how many/much resources a given app uses.

Ideas? :confused:
 
marty9876 said:
Are you running graphical client (pretty pictures) or the console service? Use the service if your not, very few problems there. Check the FAQ's here for more info.
Click to expand...

With either of them, I either get full system lockups whenever I try and game or get booted back to the desktop. Even having gone through the FAQ I was unable to resolve my problem. The simplest solution seems to be to just turn off F@H when gaming, and turn it back on when done.
 
BillR said:
I have spoken on this subject before. Stanford for some time now has and continues to exhibit a total disregard for where all there work gets done.

This project started out with the intent of using your spare CPU cycles to accomplish a goal. Many of us went way beyond that and bought, borrowed, borged many machines in hopes of finding “the cure”.

As time goes on it seems the brains behind the project have made huge assumptions about the computers we run. Yes, many of us built and continue to build faster and faster machines and whether it was for the points or sake of the project didn’t matter, we just did it and were happy to do it.

Now we are faced with work units that require real “Work Station Grade” machines. From where does Stanford think these machines will come? Certainly not the average home user. Stanford itself can’t provide any number of this type machine, what makes them think we will, or should?

Somewhere pure arrogance has set in at Stanford. I find it pretty typical of today’s colleges and schools (I’ll hold off on any liberal speeches.). They want and expect to get it all, but always at someone else’s expense.

Will I stop folding, no. I really believe in the project. Do I think Stanford needs a bit of an attitude adjustment, absolutely.

Click to expand...



..And this is my biggest complaint about this project and I've been moaning for a couple of years now.

It seems Stanford just expects us to keep ramping up to keep up with them. Especially if you want to be sort of competitive. I've walked away from this project twice now, and like Bill states, I believe in it so I keep coming back. There has to be a point where Stanford will recognize the ones that do its work and be more aware of the philanthropy involved. I think its a bit selfish of them to hog our resources so deliberately, especially when it was not suppose to be this way

As some have mentioned, my primary rig, I nice healthy xp-m 2400 at 2.4ghz with 1 ghz of ram, a nice and practicle OC of a Ti4800 wont play NFS2 when the clients are running. I suspect its a lack of RAM, and I dont think its that fair that I have to increase my memory so i can still fold AND play a game. So I go to 2gb of Memory, at my expense of course, will Stanford then just send me an UBER HUGE unit and I end up in the same boat, again ?

I wonder how many personal PC's Vijay has cooking at home, on his own personal electric bill, and what level of build they are...
 
I've run into this problem before, and lost a few borgs to it as well, and it is a bit upsetting to see it get worse. It hasn't been a problem yet on my parent's P4, but if it ever does, even once, it'll have to get knocked back from big WU's, although it'll take a while for that. The worst part is, is that a lot of users that wouldn't really mind upgrading to 2 gigs of RAM for games and such may not have the option to do so without nuking their speed. I for one would upgrade to 2 gigs from my current 1 in my main rig, but to do so would require 2 1 gig sticks of RAM, and would also force me to clock my RAM back from DDR400 to DDR333, or I could save about 150 bucks and have to clock back to DDR200. Not a very nice picture, but it's happening. What Stanford doesn't quite seem to realize is that for most people, RAM speed makes a difference more than quantity, and to get the quantities they're starting to get up to to fold requires a drop in speed, added with the fact that some people just can't afford it, or would rather put the extra money in some other part of a machine, like a faster processor, which will fold faster. Plus, a lot of people out there are switching to laptops. How many laptops do you know of that can get to even 1 gig without breaking the bank? I remember 2 years ago when 512MB was reccomended to run a big WU without noticing any performance hit, now it's getting to where a big WU will take 512MB and force a system with that much to swap. All I have to say is, the day my rig gets a WU that uses 512MB, it's getting pulled from big WU's. And, considering that "good" RAM, not even ultra-overclockable RAM, just stuff with tight timings and a good speed, is still 50 bucks for a stick of 512, I don't plan on upgrading. It may not seem like much, but with how much people already donate in terms of extra power, hard drive wear, and full system costs as well as maintenance time, it's just pushing it too far.
 
Well Vijay has finally publicly addressed the issue:
Last post here:
http://forums.2cpu.com/showthread.php?threadid=66609&perpage=30&pagenumber=1


Hi,

I hope you don't mind me visiting. I just wanted to make a brief reply to inform you of our progress relating to some of the issues raised in this thread. We proposed a fix when this initially came out and have been continuing to work on a more mature fix since then.

A lot of the discussion with Jim has occurred on the Mod side and the public hasn't been privy to my comments there. I'll post some summary of them below.

- There was a proposed fix to this issue announced once the issue was brought up 6 months ago (I've reposted the info here
http://forum.folding-community.org/...pic.php?t=12749
since I think it's been lost in all the threads). Basically, beta clients + big WU = potentially buggy, big WUs, so clients which can handle big WUs, but not QMD, etc, should just run with the big WU flag only (no beta or adv).

Granted this is a hack, but a hack that works quite well actually (and really isn't much more of a hack than the memory switch requested).

- we are working on improving the client such that these issues do not become even more of a problem (and to avoid the hack suggested above). Those more rigorous fixes are going slowly, but they are being actively worked on.

- Also in the works is a true SMP client (which really means an SMP core, which itself is a major feat which we've been working on for nearly 2 years). The clearspeed and GPU ports get the most excitement, but actually, not the most time. The SMP port has been going on for a long time, whereas the GPUs are new (and largely handled by our collaborators) and Clearspeed is largely handling their port themselves.

- we are really grateful for all of Jim's contributions and glad that he's still willing to stay around to help others.

I hope that gives some more info. 2CPU.com is a great team and we're hoping to be able to support SMP machines much better in the near future. Thanks to all of you for your contributions!


Vijay
Click to expand...

Not sure how much it well help though since it seems at least to me to be a "we are working on it".
 
You must log in or register to reply here.
Back
Top