QMDs suspended

[ChasR]

This is a KNEE JERK reaction to one large folders complaint. Nuts.

There have been some questions about the use of flags and FAH's stability and I wanted to make an announcement to try to clear things up.

There are two flags in particular which are to be used with caution. The big WU flag gets big WUs, but these big WUs may be poorly suited for older machines. The bigWU flag combined with advanced methods brings big WUs that are more advanced (QMD"s right now).

Gromacs was originally released as an advanced only WU until we could work things out to make it more stable. By having a flag, one would have to opt in and the default (Tinker in that case) would be stable.

We've kept that philosophy throughout FAH. Right now, Gromacs is stable, but QMD is much less so. QMD is also a big WU, so one has to choose both QMD and advanced methods to get it. BigWU + adv gets highly experimental QMD WUs and they may not work well for all. We suggest that one only use both flags if you're absolutely sure this is what you want (if you're not sure what you want, I would suggest that you don't want QMD's).

We should have been more clear about this, but hopefully this post will help. Due to this confusion, I'm going to suspend QMD's until we can be sure that a significant number of people have gotten the word on this.
So, as a summary, advanced methods is just that -- something which may bring more points, but also more trouble. Please consider this tradeoff when you choose settings.


_________________
Professor Vijay S. Pande, PhD
Director, Folding@Home Distributed Computing Project, and
Assistant Professor of Chemistry and of Structural Biology, Stanford University

[dicecca112]

wait wait wait, does that mean no QMDS period or I need to have advanced methods on to get them?

[ChasR]

It means period. You've always had to have -advmethods on to get them.

[dicecca112]

I have adv methods and bigpackets on so I'll continue getting them or no?

[veryhumid]

what a bad idea. That knee jerk reaction is months late.

[ChasR]

I have adv methods and bigpackets on so I'll continue getting them or no?

No.

[Audioaficionado]

I guess it's time to fire up all 6 of my HT clients again once the QMDs have all cleared. I left them all installed just for a day like this. I'll remove that -one unit flag and put all the advanced/beta flags on them so I'll get the latest and greatest other than QMDs.

Instead of fixing the problem, they're going to spank us all for even a lot of constructive critisism. OC-AMD still won't come back until they actually fix it. He's tired of bandaids. I understand his POV as he can't futz with every single system like most of us can.

When I see QMDs back, I'll drop back to my usual 3 Intel clients.

[Silver]

How's about a link so that those of us who are capable can voice our sentiments where it might count.

Obviously what we need is a -QMD flag.

[Audioaficionado]

http://forum.folding-community.org/viewtopic.php?t=12753&sid=3fd45dd962986af1d4cbaf6ea9ccc16f

[ChasR]

QMD Suspended thread:

http://forum.folding-community.org/viewtopic.php?t=12756

[AMD_OCER]

I don't know what you guys think about this, but I think It sucks. :mad:

I know my PPD went up after switching three of my P4's to QMD's. :cry:

I am going to miss them and can't wait until they come back.

[ChasR]

I guess it's time to fire up all 6 of my HT clients again once the QMDs have all cleared. I left them all installed just for a day like this. I'll remove that -one unit flag and put all the advanced/beta flags on them so I'll get the latest and greatest other than QMDs.

Instead of fixing the problem, they're going to spank us all for even a lot of constructive critisism. OC-AMD still won't come back until they actually fix it. He's tired of bandaids. I understand his POV as he can't futz with every single system like most of us can.

When I see QMDs back, I'll drop back to my usual 3 Intel clients.

I'm set to go back to HT as well. As I've said before nothing has changed in the publicly released WUs. The entire crisis is in reaction to a WU that may never see the light of day.

[Audioaficionado]

Heck with my 2gigs of RAM I could even handle a couple of those bad boys :D

[muddocktor]

Heh, instead of fixing the damn problem Vijay just suspends QMD's, how typical of Stanford. :rolleyes:

Thank God my farm only had 2 P4 machines so I won't take as big a points hit as other folks. :rolleyes:

[Audioaficionado]

They should be back in a week at most according to Vijay.

[OC-AMD]

I wouldn't call this a knee jerk reaction.. in the conversations I've been having with Vijay, I raised the point that his recommended fix of not running certain combinations of setting/flags is only good to those who read his recommendation at the community forums... I can tell you only the very dedicated folders such as those here and other teams are members there, there are many, many more users that don't know about the forums... but run -advmethods because they heard it will give them more points without knowing what it's really doing.

Suspending the unit's really isn't going to solve anything until the client says what the heck checking "large WU" actually means... it was originally suppuse to mean a file download of 5Mb or more... it got twisted into meaning it may use a heck of a lot of RAM. (and if you've seen some of my posts, I had 3 sessions of F@H using 2.4GB's of ram on a single server)

And while QMD's are part of the issue on a dually, it was a also the lack of memory vs CPU detection.

[portorock]

what is the big wu flag? I know what it does, but I want to know what to put so that when they do start them back up I can get them

[TollhouseFrank]

hmm.... I hope that this gets solved soon, and that the QMD can come to us A64 guys....

[Silver]

hmm.... I hope that this gets solved soon, and that the QMD can come to us A64 guys....

Second that however I am inclined to pull the plug now as well till things get fixed. Tired of sending units in and getting no credit. The Celly is folding a QMD in 24 hours but stanford won't give it to the celly. Had to down load them to the p4 then move them to the celly and return them to the p4 to send in. The a64 should have had qmds a while back, why are we still waiting? why do I have to move units from p4s to a64s in order to get optimum production? As for the ability to use the systems effectively for other things, well we know that is not so. I'm still folding as of this minute, however I am really not convinced of the integrity of the people running this project. Efficiency seems to be the last thing on their minds and our cpu cycles, parts and time, paid for by us and our families are not theirs to waste. If they can not use our resources effectively then why should we let them be used? If there is a single big folder on team 32 who stops in protest then I too will pull the plug. It seems that the only really effective way to get their attention and to ultimately increase their results and thereby the intent of the program is to make a stand. It would appear that Stanford is unwilling, does not have the time, is lacking in expertise or just does not have the funding to provide how many free cycles to aid in its' research. This is an intolerable situation given the amount of money spent by the members that are folding to help what we percieve as mankind.
Though it is a ruthless means, if there is enough of a drop, Stanford will immediately address the situation or have to concede that our contributions are of little interest to them and our efforts would be better served elsewhere.

http://forum.folding-community.org/...4cbaf6ea9ccc16f

Can't imagine why they locked that thread.

Xeons are presently relegated to folding 4 Tinkers. The P4 will be in the same boat tommorow (actually a couple hours). Then I will go back to hyper threading it and may down clock it some as whats the use. As for the a64s and the celly, may bring them down all together tommorow.

[ChasR]

I wouldn't call this a knee jerk reaction.. in the conversations I've been having with Vijay, I raised the point that his recommended fix of not running certain combinations of setting/flags is only good to those who read his recommendation at the community forums... I can tell you only the very dedicated folders such as those here and other teams are members there, there are many, many more users that don't know about the forums... but run -advmethods because they heard it will give them more points without knowing what it's really doing.

Point taken. To assume everyone pays FAH as much attention as those on this forum, is unwise. When I stop and consider how many borgs I lost when QMDs slowed them intollerably and how that would also apply to every preson downloading FAH for the first time, it gives me pause. I don't think suspending QMDs is the right way to go about it since, for the most part the damage is done, and the suspension will cause a firestorm of its own. Far and away the most prevalent problems are associated with multiple instances on a single machine and you pretty much have to visit the forums to find out how to do that.

[ghettocomp]

I hope those WUs will be back soon, I have 5 P4 rigs ready to borg for those being returned from maintenance soon.

[consumer9000]

Wow...lame. I've been enjoying running 450+ PPD workunits since QMD's hit the scene! WTF...now I'm getting 51PPD Gromac core crap. DOH!

[MLMIB]

my complaint, and main worry is that I can't access most of my borgs in sort notice to react to unstable WU's. I just wish they'd come out with a more intelligent client so that way I wouldn't hafta fuss with all of this back 'n forth batter....

[Audioaficionado]

Just sent out my last QMD :cry:

Now I'm running

3x p1477_release31_0 364 point Gromac
3x p1479_tet_1479_1 364 point Gramac

No tinkers at least :-/

I don't know what my new PPD will be but I was at ~1200 PPD running 3x QMDs & a couple AMD boxen set for -advmethod & >5MB BIG WU to try and snag the occasional Big Packet Gromac.

I'm back to running my intel processors in HT mode to try and eke out the maximum PPD until QMDs come back.

[Bon3thugz43v3r]

Well, personally I think it's funny that everyone is complaining, afterall people seem to preach that it's not all about the points, its about the research. Yet, when QMD's (aka point boosters for p4's) get shutdown, everyone complains. I don't have a problem with QMD's getting shutdown at all. I remember when folding used no more then 20-30 megs of memory, now its becoming out of hand. I don't know how others feel about the memory usage, but personally I think its bs. Folding used to not affect performance, but now it definitley does.

[Audioaficionado]

Yeah you have to build a dedicated folder anymore if you want the big wu action. If you want to just help out in the background, I think those timeless WUs shouldn't suck up the resources too bad.

If you don't want the QMDs you just have to adjust your settings and you shouldn't get them anymore. Same goes for the Big Packet Gromacs.

Tinkers and small timeless gromacs are hardly noticable in the background.

Hopefully the assignment server remembers which machines were reliable QMD folders and resumes the QMD stream back to my three Intel processors.

[psyshack]

Looks like I will be turning in my last QMD here in a few hours. :(

But Im sure boinc research will move right along. :)

I have to agree with Silvers post up there ^^^^^^

[Audioaficionado]

It looks like I'm getting those popcorn kernal 51 point gromacs as a QMD replacement. Lets hope the assignment servers remember who had the good QMD machines when they resume them.

[muddocktor]

Heh, on my 2 P4 machines I presently have a p147x and a Tinker on one and the other drew 2 p147x wu's. I'm presently experimenting with the one with the 2 p147x wu's to see if it gets better production running 1 client or 2 with these wu's. It doesn't even come close to my mA64 folders and the P-M laptop though with these wu's, even running just 1 client.

[Audioaficionado]

p1477_release31_0 364 point Gromac
p1479_tet_1479_1 364 point Gramac

My son's single 2.2 GHz Barton 3200 is beating the 3.0 P4c and matching the 3.3 Xeon duallie on these.

So it appears Stanford's WUs are all over the map when it comes to which folds better on what processor.

[Macaholic]

Looks like the QMDs will be back on Monday, BIG caveat for those with dedicated folders though;

Posted: Today at 7:41 pm Post subject: UPDATE
Our plan is to re-release QMD WUs starting on Monday ~noon PST. Hopefully enough people will have heard about the various issues. We will release them gradually, starting with a high memory limit. The previous memory limit was set to 500MB. We'll start conservatively with a limit of 1000MB and then gradually work down from there. Note that in the current client, the memory reported to the AS is the total memory of the box. For SMP boxes, all clients will report the same amount of RAM, which is bad since they are in fact all sharing that RAM. Thus, we will re-release these with a large RAM request to try to help this issue.

In addition to this re-release plan, we are working to release a new v5 client rev which allows the donor to set the amount of RAM reported to the AS. Thus, if your dual has 1GB and you are running 2 clients, you can choose to report to the AS that each client should only use 500MB. One can also use this to simply request smaller WUs on machines. This is a "soft limit" as there is no code to ensure that the core sticks to this, but this soft request will still go a long way to allocating WUs correctly to machines. We are also considering the possibililty of a hard limit, which would stop cores from using more memory than allocated. It is unclear whether this is a good idea (and how to do this right) and we will do more testing on this before we release this feature in v6.
_________________
Professor Vijay S. Pande, PhD
Director, Folding@Home Distributed Computing Project, and
Assistant Professor of Chemistry and of Structural Biology, Stanford University

I don't know about some of you, but I've got several dedicated P4s that had been running QMDs just fine with only 512MB RAM. But that won't happen any more since the limit has been raised to 1000MB RAM. Full thread here (http://forum.folding-community.org/viewtopic.php?p=106478#106478).

[KingFish]

Wow, that's a pretty fast re-release.

[Audioaficionado]

Yeah I'm lucky to have an extra 1GB DC kit laying around or I'd lose the P4c Shuttle with only 512MB of memory on it. I've got to install that tonight :(

[muddocktor]

Well, I'm in good shape for this at least because both my P4's already have a gig of ram in them. Be glad to see them come back.

[iceage]

I wonder how that will affect alot of the borgs that are out there. Being that I see alot of them with 512 or 256 meg ram.

I'm glad I have 1G of ram in my dualies. ;)

[Silver]

Well for me it means that two of the three p4s won't be doing qmds any more. Been running qmds just fine with 512. I'll be darned if I am going to spend a few hundred more for ram in order to be given the honor of running qmds for them. so much for the rant. Not really into little green monsters so anyone know of a really good research program other than FAH?

Edit, in order to acquire the qmd 1000mb is required but that should make no difference in the actual running. One should be able to download and pass these to 512mb machines for processing. Using the -oneunit flag one could just go to the next. I would believe this to be correct as I did some testing on the a64 w/256mb ram and qmds as well as passing work to the celly. Anyone see any reason why this would not work? Results should upload just fine as completed units.

[dwschoon]

That really sucks. None of my machines has more than 512mb ram. I was really enjoying the qmd's.

[muddocktor]

That should work just fine, Silver. If I were you, I would just run the client with either the -one unit or -pause flag after you've transferred the wu('s) to the machine and then load another couple of QMD's to it. I would also run a couple of clients to keep from having to transfer the work as often too. You shouldn't have any problems with sending the results back with the machines though, since the server doesn't care how much ram the machine has that's sending the results in from.

[Silver]

Run
msconfig
Uncheck the fah clients
reboot
create a couple folders on the desktop with the names fah1, fah2, etc.
download the files from your email
put them into the desktop folder you want them to run from
edit the client file with your name
click on console and make a shortcut
click on short cut properties and add the flag -oneunit (this will cause it to stop after sending the results back and not download another.)
When it is done just go to the next folder and start that one up.
Always start it with the shortcut you created in order that it sees the -oneunit flag and does not auto download a client you have not intention of running. IMPORTANT.

Anyone (within reason) that has only 512mb of ram and wishes to recieve qmds then either pm me or email me with your e-mail address and how many you would like. I will not send more than four to anyone and ONLY team 32 members. These have to be manually downloaded, zipped and sent to another site for your downloading so a touch of patience is needed. Do check your e-mail given as it will notify you when the file is ready for download.

Edit: time is 1:33 PM (Georgia). Is anyone getting a new QMD yet?

[muddocktor]

Just checked the server stats page and it's still showing server 113 as "Accept" instead of "Full", so I imagine they haven't gotten it modified for the 1 gig memory check yet.

[dwschoon]

I just had a machine send in a qmd and in return it got p1479 which is a gromacs. it is a celly D 330 w/ 512mb of ram. I wonder, will the new client allow you to tell the server you have more ram than you actually do? or will it only allow you to decreas the ammount reported.

[pik4chu]

woke up this morning to find my 2 600 point QMDs had finished and sent in and were replaced by a 51 point gromac and 67 point tinker.....And thats with my xeons running QMDs almost non-stop for a few weeks now =\ ...Well time to power down some rigs, the heat output is not worth this, good cause or not, sadly. Yep its not just for the points, but when I have to sit in a room thats pushing 90F because my A/C broke, staring at this pidly little tinker as the source of my misery, just can't handle that anymore.

As for RAM hogs.. I don't (and never did) have a problem with the bigpackets thing. Heck on main rig I run 2 instances of FAH and they can run 2 QMDs non-stop (chewing ~256MB ram each) AND play BF2 @high detail w/o a single issue. so :shrug: never knew there was any problem with them myself.

[Silver]

My dually is down as well. I have a little work in the ques of the p4 and celly but if QMDs are not available by then they will be going down as well. So far over 6Ghz is off line with another 7Ghz coming off soon. Everything is 50 and 60 pointers for me as well and I am not going to run 24/7 for a measley 150 points. What a waste of processing power. P4 will go down tonight. So almost 11 Ghz down as of today.

[Silver]

QMDs are up again....if you have 1gb of ram.

[dwschoon]

I think the vast majority of people who want qmd's dont have 1gb of ram. I am not about to spend the money to upgrade my computers. I have a p4 right now with 384, and i am waiting on a 512 stick in the mail. I bought it expressly to get qmd's and now that isnt going to happen even then. My highest point producer went from 250ppd on qmd's to 160ppd on gromacs. This sucks.

[Audioaficionado]

Well I have 3 fully qualified processors that have all the advanced QMD configs and flags set. They get Tinkers and 51 point Gromacs now.

The two AMD processors get the BigPacket Gromacs.

Like OC-AMD and others have said, flags and settings don't always work. You'll get whatever that server has on que at the time your client asks for new work.

I'll give 'em till the end of the month to straighten this mess up. After that I may pull the Intel processors and only fold with the two AMD XP processors. I won't waste that hand picked advanced folding hardware on improperly assigned work units.

Hopefully things will settle down shortly and we can get on with it.

[muddocktor]

The QMD server (server 113) is back online!!!

I just d/l'ed my first QMD a few minutes ago. You should be able to get them now with machines configured with 1 gig of ram and the -advmethods flag set.

[pik4chu]

lame, any chance of faking the 1gb of ram? not gonna put an extra stick in a fully capable machine just because they cant get thier stuff straight.... :mad:

[nikhsub1]

Me too me too me too.

[20:43:35] Project: 1912 (Run 248, Clone 0, Gen 43)
[20:43:35]
[20:43:35] Writing local files
[20:43:35] Extra SSE2/SSE3 boost OK.
[20:43:35] Entering QMD...
[20:44:04] System: p1912_ALA-dipeptide_umb2
[20:44:04]
[20:44:04] Performing initial WF calculations
[20:44:04] - Number of total steps will change until convergence
[20:44:39] Completed 0 out of 2000 steps (0)
[20:52:53] Completed 21 out of 2021 steps (1)
[20:52:53] Writing local files

[Joe Camel]

bah, just when i thought my rigs / production (most "older" Celly / AMD / PIII) would start looking "good" again.... :rolleyes:

[ChasR]

Got one and can't wait to fill up again. Sad to say, my P4 1.4 and 1.5 are unlikely to ever fold them again unless I can force myself to spend the $ on 512MB of RDRAM for each.

Edit:
Its worse than I thought as both are populated with 4 - 128MB sticks. It'll take totally new ram. No Way.

[nikhsub1]

Stanford needs to have a workaround flag for this. The default should be 1GB of ram otherwise no BP/QMD, but, there should now be a flag to specifically allow BP/QMD on machines with lesser ram. This is how it should have always been. Is this in the works?

[nikhsub1]

Just slap me... the new release will have this.

[Audioaficionado]

I was finally in the top 50 when this all started but now I'm back to #51 and soon to be #52.

It wasn't Stanford's fault LOL. I'm just getting stomped into the second 50 mud by two 3k+PPD Red folding monsters.

[AlabamaCajun]

I think the QMDs are out there, I just got a 1912 -ALA-Peptide Core 96. :D

About the memory requirements, that should drop after a while according to the F&H forum.

[ChasR]

woke up this morning to find my 2 600 point QMDs had finished and sent in and were replaced by a 51 point gromac and 67 point tinker.....

600 pointers aren't QMDs they're big Gromacs. QMDs are 450 or 125 points.

[mjw21a]

Hmmmm, I think this might actually benefit me when competing for points as II don't really have many P4 rigs, and only 1 with 512MB RAM or more.