New to F@H need startup info

This client will only use a single CPU

Moderators: Site Moderators, PandeGroup

Re: New to F@H need startup info

Postby uncle fuzzy » Sun Dec 20, 2009 12:34 am

Just stick with it. Posting this thread was a good start. Any time we throw out something you don't understand, just ask. As you've seen, those who have been here a while can interpret for each other. We all speak "folding", no matter where we come from.

For a little light reading, here's the guide I used to start running notfred's.
http://www.overclockers.com/forums/show ... p?t=599272
There are many out there. You can Google notfred smp vm.
Proud to crash my machines as a Beta Tester!

Image
User avatar
uncle fuzzy
 
Posts: 978
Joined: Sun Dec 02, 2007 10:15 pm
Location: Michigan

Re: New to F@H need startup info

Postby jrweiss » Sun Dec 20, 2009 12:53 am

MtM wrote:
jrweiss wrote:To clarify, when MtM talks about "contribution," he means the "Points per Day" or PPD count that you see bandied about these forums a lot. While it is not a strict measure of performance or value to the Project, it's the only quantitative measure we have as contributors.


Actually I ment scientific contribution, while it's maybe not fair to say so, but smp does have a higher scientific value in the long run...

Am I wrong?

But yeah, on the other hand, it's easy to use ppd as measurement, though I really don't look at it that way. I know people who have very high ppd, but are running two vm's on each quadcore. In my opinion, even OP contribution is higher as he is not intentionally introducing delays. So, no I don't talk about PPD as 'contribution', I thought we been through this before ;)

OTOOH, how do we quantitatively measure "scientific contribution" in any manner at all, other than using the Pande Group's own point system? It's not perfect, but it's all us commoners have. While the point count for CPU WUs may in some cases be 'artificially inflated' to give those with lesser-performing machines more incentive to Fold, that only illustrates that PG NEEDS those WUs as much as the SMP and GPU WUs.

I understand your point of view. In fact, in a roundabout way it agrees with mine, because I also noted that the SMP client will yield more points in the long run. For the newbie, PPD is the only way to begin to understand the relative scientific value of the various WUs (Work Units).

AND, I totally agree with your point of view that those who are forcing a few more PPD by going against the Project's recommendations of 1 client per physical core and introducing delays in individual WUs DO ultimately produce a "lesser" contribution by subjective criteria, even if their PPD seems to indicate otherwise.

Yes, we've been through this several times, but it's worthwhile because it's not a discussion a newbie will likely target in an archive search...
i4770K/MSI GTX 1050ti
i7-3770S/XFX RX460 fanless
User avatar
jrweiss
 
Posts: 1100
Joined: Tue Dec 04, 2007 6:56 am
Location: @Home

Re: New to F@H need startup info

Postby codysluder » Sun Dec 20, 2009 2:46 am

jrweiss wrote:AND, I totally agree with your point of view that those who are forcing a few more PPD by going against the Project's recommendations of 1 client per physical core and introducing delays in individual WUs DO ultimately produce a "lesser" contribution by subjective criteria, even if their PPD seems to indicate otherwise.


Things may change soon. If the TRIAL bonus system is adopted by FAH, then the points system will essentially change. The project recommendations will coincide with a maximized PPD. Those who presently introduce delays will stop doing it because that will cost them in terms of a reduced total PPD. (at least that's my assessment of the trial bonus system.)
codysluder
 
Posts: 2128
Joined: Sun Dec 02, 2007 12:43 pm

Re: New to F@H need startup info

Postby MtM » Sun Dec 20, 2009 9:34 am

uncle fuzzy wrote:
MtM wrote:I know people who have very high ppd, but are running two vm's on each quadcore. In my opinion, even OP contribution is higher as he is not intentionally introducing delays. So, no I don't talk about PPD as 'contribution', I thought we been through this before ;)

This does not always mean slowing down the science. I run Intel quads- 3 Q6600s and a Q9550. On all 4 of them, running a 2-core notfred's VM resulted in higher PPD, and a quicker completion and return, than running a Win SMP on all 4 cores. Running 2 2-core VMs allowed me to more than double the output of my CPUs. Now that we have Player 3, a 4-core VM gives a slight advantage over a pair of 2-core.

Basic math- PPD is proportional to speed of returning the results, therefore, VMs do not slow down the science (depending on hardware). YMMV.


Wrong. Read more, talk less please.

@Codysluder, thumb up :) (edit: sorry misspelled your name there :oops: )

@jrweiss, good points and as always glad to have you around.
Last edited by MtM on Sun Dec 20, 2009 4:49 pm, edited 1 time in total.
MtM
 
Posts: 3054
Joined: Fri Jun 27, 2008 2:20 pm
Location: The Netherlands

Re: New to F@H need startup info

Postby uncle fuzzy » Sun Dec 20, 2009 4:45 pm

Where am I wrong if I double my points AND return the results quicker?
User avatar
uncle fuzzy
 
Posts: 978
Joined: Sun Dec 02, 2007 10:15 pm
Location: Michigan

Re: New to F@H need startup info

Postby MtM » Sun Dec 20, 2009 4:55 pm

Uncle fuzzy you're a beta tester even, how can you even ask me this :e?:

I'm not joking if I say I'm surprised you do not know, and tbh also abit shocked and dissapointed you asked me instead of used the forums search function.

I been involved in way to many explanations already, I feel it's getting real real tiresome to keep repeating it, or looking up the previous discussion for others who are usually new donors. Imagine how it feels to be told by a beta tester, from all persons, who has concluded without research I assume that...

This does not always mean slowing down the science. I run Intel quads- 3 Q6600s and a Q9550. On all 4 of them, running a 2-core notfred's VM resulted in higher PPD, and a quicker completion and return, than running a Win SMP on all 4 cores. Running 2 2-core VMs allowed me to more than double the output of my CPUs.


Please search for some info on your own, when you done so and want to comment on those discussion I'm more then willing, eager even, to participate in any meaningfull discussion about it, but don't ask me to hand it on a platter ( not today at least, you got me on a bad moment I confess as I got a headache the size of Everest :( ).

Edit: spelling, I wasn't joking about the headache!
MtM
 
Posts: 3054
Joined: Fri Jun 27, 2008 2:20 pm
Location: The Netherlands

Re: New to F@H need startup info

Postby rtbrd » Sun Dec 20, 2009 7:34 pm

In preperation for deciding which way to go, multiple uniprocessors or SMP, would I be correct in assuming that a higher PPD would indicate a higher priority/importance of the WU to the researchers? I guess my problem is that I don't know how work units are assigned. I fact I'm really not sure what I am doing. I work in electornics and have little to no knowledge of medicine. Are WUs assigned to specific clients due to their complexity or are you just assigned the WU that is next in line.

I have run 5 WUs so far and I am working the next 4. I have been monitoring progress and have started using seconds/step as a guage to get an idea of how fast things move. With the exception of two WUs I am averaging ~120ms per step based on the data provided in the log file. If anyone has a better idea of how to monitor progress please let me know. Using the guage I have set up I can see "real time" results knowing what other things I am doing on my computer.

Another question. I can see how many bytes are downloaded for a WU as reported in the log file, however and am not sure of the upload burden. The reason for this question is that I have some friends that I would like to recruit but they use wireless for connection to their ISP (I use copper) and they are concerned about exceeding their free bytes/month. Knowing both upload and download burden and average time to complete WUs I can calulate the needed volume and maybe get them into the fold, pun intended.
Last edited by rtbrd on Sun Dec 20, 2009 8:02 pm, edited 1 time in total.
rtbrd
 
Posts: 26
Joined: Wed Dec 16, 2009 9:32 pm
Location: St. Petersburg, FL

Re: New to F@H need startup info

Postby MtM » Sun Dec 20, 2009 7:53 pm

Yes.

@ Uncle_Fuzzy, you are right, 2 vm's on a windows box is more beneficial then a single windows smp client if you return them quicker, but I tested that myself extensively and I find it to be not an accurate representation of performance. The windows A1 core is not slower on 4 cores then the A2 core is on two cores, it's infact about 10% quicker on average as I gatherd from own testing.

There is a strain of A1 projects running, all on windows machines. This is line a.

There is a strain of A2 projects running, all on linux/osX. This is line b.

When you remove a quad which was in line a, and change it to a dual vm which do work for line b, you remove a client running as intended ( one fahcore per physical core ) eg it is offering the best speed obtainable for that production line.

You exchange this 'most beneficial' setup with two vm's which each on their own are running in an unintended way. Smp was meant for quads, not dual cores, and it has always been that way. But yes, Stanford accepts donations from dual cores as they welcome any help they can get, this does not mean it's as beneficial to them as clients running in the preferred manner.

Let's look again at line b, imagine a long line of quads working on that line, and now you add two dual cores. These dual cores are slower then the quads, and when the dual core is busy working on the wu the entire line of wu's behind it get delayed when compared with the preferred configurations running the client as intended. It's been explained by people way more qualified that the serial nature of f@h suffers from this delay allot, enough to make the scientific contribution you're making worth less then the points you get in this scenario. So, if you're taking ppd as measurement, PG has already stated many times this is incorrect. The bigadv bonus plan should give a hint of where we are going in the future with quick return resulting in higher point bonuses.

Both statements are therefore incorrect.
MtM
 
Posts: 3054
Joined: Fri Jun 27, 2008 2:20 pm
Location: The Netherlands

Re: New to F@H need startup info

Postby jrweiss » Sun Dec 20, 2009 8:00 pm

You will not find PPD listed in the Project or WU -- only total points credited for each WU. The PPD count depends on your individual system specs and setup, and is normally gotten from a 3rd party utility like Fahmon.

The points credited for a particular WU are based on the Project's assessment of the time required to complete the WU on their baseline machine (not sure what it is right now). They are balanced to give approximately the same PPD for a "normal" WU on that machine regardless of the WU involved. However, AMD CPUs crunch different WUs (especially depending on the core -- Amber, Gromacs, Double Gromacs, etc -- used for that project) faster or slower than Intel CPUs, so the PPD of any individual WU on a particular machine may vary widely.

ALSO, when the Project managers decide a certain project or group of WUs is especially important during a certain time period, they will award bonus points as an incentive for those whose machines are capable. These are often in the "Advanced Methods" category that you choose to Fold or not during the config process.

So, after all that, the point system is designed such that over the average of all WUs, projects, and computer setups under supported configurations, a higher PPD signifies a higher "contribution" or "importance." Because of all the variations, there is no way for you to tell in advance what particular configuration will yield the highest PPD on any particular day. In general, though, over the long term a stable quad-core system with a current nVidia or ATI GPU will yield the most PPD over the long term with an SMP+GPU client configuration. Those adventurous Folders who have a 64-bit OS with plenty of RAM will usually see more with SMP+GPU under Windows, plus another SMP client in a Linux Virtual Machine.

Things are likely to change with new clients developed to make better use of new CPU (e.g., i7) and GPU technologies.
User avatar
jrweiss
 
Posts: 1100
Joined: Tue Dec 04, 2007 6:56 am
Location: @Home

Re: New to F@H need startup info

Postby rtbrd » Sun Dec 20, 2009 8:10 pm

Boy, yall answered fast. I was in the middle of editing that post when yours came in.
rtbrd
 
Posts: 26
Joined: Wed Dec 16, 2009 9:32 pm
Location: St. Petersburg, FL

Re: New to F@H need startup info

Postby MtM » Sun Dec 20, 2009 10:40 pm

Sorry, the 'yes' only appleid to the first sentence ( which was all which was there atm ) ;)

If you want to measure your productivity, you can use any of the popular tools listed in the 3rd party contributions section of the forum. I am a big fan of HFM.net, I would advice to use that but there are others and the choice is yours to make. That way you don't have to go through the logs or look at the console windows ( infact, if they are running as intented and if you want to keep them for the time being, I would exit the client with ctrl-c, run the configuration again with -configonly ( explained in the guides if you don't know how ) and configure each client as a service. Then go to the start menu -> run -> services.msc and look for the Folding@Home service listings ( should be one for each client ). Start the services from the control window and you're folding without the 'nasty' consoles on your desktop.

How do you get assigned work units? Atm this only is influenced by the configuration settings, not the hardware ( exception is gpu as the client checks board vendor as Ati and Nvidia use diffrent cores ). Cpu's, wether Amd/Intel/Cyrix ( do they still exist??) use the same core, and while some cores are more geared towards a certain architecture ( sse variants ) they do not differentiate on this potential during assignment. If you opt for small wu's, and not use advanced settings, you will get work units with a small footprint ( network usage, will come back to that later on ) and projects which have been running for awhile usually. If you say yes to advanced methods, you opt to run small work units which have just left beta testing and can be abit more unstable but also offer a point benefit most of the time ( all of time? not sure and I got an headache, if I'm wrong I will get corrected ). I say opt, as it's up to the assignment server to determine what you will get, and while you opt for something there might be no work units available with your specified characteristics. If so, it will output to the log that no appropriate work can be found and you have to look at your settings and make some adjustments.

The exact order of importance for each flag is not easy to give, I think even on a good day I couldn't say for sure and this is by far one of my good days ;) There are 2 settings which are most influential and those I already mentioned, the work unit size, and advmethods.

Now network usage, small wu's are 1.5-2 mb iirc, but I must add I am not sure as the size descriptions have changed with the transition from the v5 clients to the v6 clients, and the documentation on fahwiki about network usage I think still refers to the old client settings ( big/small ) while now it's small-normal-big.

It is safe to say big will strain your connection more then normal offcourse, and for anyone with a bandwidth cap I would advice that. On the other hand, there are people on the forum who know exactly what each client averages in bandwidth ( I don't as it never was any concern to me ) maybe one will see your question and give it a better answer then I can.

If their connection to their isp is wireless, I would worry about connectivity loss, the clientcode doesn't deal well with disconnects ect. Also something to consider perhaps.

I will ask someone who I think can answer atleast some of your questions with better accuracy to look at the thread.

Now, off to get cured from this headache! ( off to bed I go ).
MtM
 
Posts: 3054
Joined: Fri Jun 27, 2008 2:20 pm
Location: The Netherlands

Re: New to F@H need startup info

Postby bruce » Sun Dec 20, 2009 11:40 pm

SOME of the charactistics of your client and your hardware influence the assignment process. In no particular order, here are the ones I'm thinking of right now:
* Which OS are you running?
* Which client did you choose?
* Did you select Big/Normal/Small?
* Did you set special flags such as -advmethods or -bigadv?
* For the GPU client, which brand do you have?
* For the SMP client, how many cores does your OS report that you have?
* If you have too many EUEs (errors) or lost WUs within a certain amount of time.

Can be an influence, but rarely important, today:
* How much RAM do you have? (and did you set a limitation during the -config/-configonly process?)
* Your recent history of the speed at which you return results

Not considered:
* brand of CPU
* MHz of CPU
* The speed of your intenet connection

(I've probably forgotten a couple of things. If I remember them later, I'll edit them in then.)

The scientific importance of various projects changes from time to time. When the servers find more than one Project that matches the criteria you've set, the Pande Group assigns them mostly a random. They can adjust the relative priority of projects either based on the importance of science or to minimize server congestion so it's not really purely random, but none of us can see enough factors to use this information since they're the one doing the assigning.

As has already been said, the base points for any project are assigned based on how long it runs on a standard machine. Historically speaking, bonuses have been added for a variety of reasons but they have limited value since the factors you can control don't relate directly to whether a project has a bonus or not, although indirectly.

Originally, points were directly related to how many CPU-hours you contributed. The TRIAL bonus system (probably the "planned" bonus system) rewards both the quatity of CPU-hours you contribute and how quickly those results are returned.

e.g.- Assume Linux. Assume a Quad. Assume the SMP client(s).
You still do have a choice of running two WUs each on half of the quad (e.g.-with VMs or multiple Linux SMP clients or ...) or you can choose to run one WU on on 4 CPU cores. Twice as much power devoted to a single WU is better for science than splitting those CPU-hours across 2 WUs. The present bonus system doesn't reward performance that way. The upcoming bonus system will, however, reward those who eliminate those pairs of 2-core VMs.
bruce
 
Posts: 21053
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: New to F@H need startup info

Postby rtbrd » Sun Dec 20, 2009 11:40 pm

Well, there is quite a bit to digest in the last three replies, talk about a headache. I intend to let things run as they are for whle just to get a feel for how it all operates. As I progress (hopefully fast) I will try this and that. As has been posted, I will in the end decide what I think is best for my particual situation and hopefully that will be the correct setup. I must say that I what want is the best for science and I will do my best to put my 30+ years of working in the electronics field to use. When it comes down to it, science is science no matter what the field and I will apply all that I have learned to do my best.

This may sound like a sign off but it is not, as a neophyte I'm sure that I will have many, many more questions.
rtbrd
 
Posts: 26
Joined: Wed Dec 16, 2009 9:32 pm
Location: St. Petersburg, FL

Re: New to F@H need startup info

Postby bruce » Sun Dec 20, 2009 11:45 pm

We did tell you (early on) that it was complex and somewhat dynamic and it's best to start simple. I don't consider it a sign-off, but rather "enough for now"

:D
bruce
 
Posts: 21053
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: New to F@H need startup info

Postby rtbrd » Mon Dec 21, 2009 12:23 am

You got it.

"complex and somewhat" is quite an understatement. :) I'll be back, probably tomorrow.
rtbrd
 
Posts: 26
Joined: Wed Dec 16, 2009 9:32 pm
Location: St. Petersburg, FL

PreviousNext

Return to Windows Classic V6.23 Client

Who is online

Users browsing this forum: No registered users and 1 guest

cron