Predictor

[excaliber]

http://predictor.scripps.edu/



Predictor@home is a world-community experiment and effort to use distributed world-wide-web volunteer resources to assemble a supercomputer able to predict protein structure from protein sequence. Our work is aimed at testing and evaluating new algorithms and methods of protein structure prediction in the context of the Sixth Biannual CASP (Critical Assessment of Techniques for Protein Structure Prediction) experiment. The goal is to utilize these approaches together with the immense computer power that can be harnessed through the internet and volunteers all over the world (you!) to address critical biomedical questions of protein-related diseases. Predictor@home is a pilot project of the Berkeley Open Infrastructure for Network Computing (BOINC)


Im running, want to see how BOINC fairs

[FoBoT]

i am in

with my laptop for now

hopefully tomorrow i can add some home boxen, now that i have a home LAN again, i'll try to rebuild my home production

:drums:

these work thingys don't take long, on this 1 Ghz P-3 i am finishing one every ~9 minutes, i wonder if i get one point for each? hmm... it hasn't sent any back yet


this boinc thing is interesting, lots of info built in

[pfb]

running it one 1 PC - seems interesting that another folding project is using BOINC (and for CASP)...

one thing that is odd is that the work progress sits at 0% then suddenly goes to 100% - maybe because it's taking ~3 minutes to do a work unit...

also interested to see how the scoring system works in BOINC...

/edit - hmm, doesn't like my duallies - keep getting Windows errors

[IronBits]

I put one boxen on it for now. :thumbs:
added in the Seti@Home project as well :)

[FoBoT]

how long does it take for returned work to turn into "granted credit" ?

is it just once a day? or is it manually done by a person at the project?

[excaliber]

Well, its not manual, as I havent done anything and it seems to have updated. My guess is that stats are updated once a day.

[pfb]

Anyone actually got worked buffered? I've set the preference to 1 to 2 days work but the client just downloads 2 or 3 workunits, process them and then sits their saying that it has no more work and can't communicate with the server...then spends around 10-20 mins waiting and the whole cycle is repeated...

very annoying...:swear:

/edit - seems there was a break in the supply of work earlier today...oh well...

[iggy]

As it states in documentation provided (if one can call it that), it can take even few days for the points to be awarded. It seems that the same work unit is sent to two nodes and then the results compared. If all matches, points are granted.

On individual profile, they list points claimed by the system(s), but not yet verified.

I'm just wondering what the quality of the results is and where can it be seen (RMSD probably), as I can't find it anywhere among the stats. :confused:

I might give it a spin later on.

One good thing is that the client (obviously open source, aas sources are provided) can be compiled for any system, so it might take into account CPU characteristics. Downloads for Linux provide optimised binaries for i686 CPUs. There is also a benchmark option - not sure if this only tests the CPU, or how well the client performs under different kernels, OSs, and/or compile options.

[FoBoT]

mine ran out of work earlier this morning

i told it to cache 2-4 days of work, but it will only get 2 or 3 work thingys at a time, which is only about 1/2 hour of work

i think it is having a problem with the estimation of how long the work will take. since it thinks one work thingy is much longer than it really is, it won't download more at a time

i guess

[pfb]

Originally posted by FoBoT
i think it is having a problem with the estimation of how long the work will take. since it thinks one work thingy is much longer than it really is, it won't download more at a time

I've noticed that as it says the units will take 23-30 hours to complete (but only takes 3-4 minutes) - I'm guessing it's using the 23-30 hours part to work out how many to download to fulfill the cache options...

[FoBoT]

is there an easy way to add SETI work to the boinc client? or do you really have to redownload the whole thing/register new with the SETI site?

[pfb]

iirc you just need to register to get the url and Account key for each project.

http://setiboinc.ssl.berkeley.edu/ap/create_account_form.php should be the link for Seti

http://wibble.bounceme.net/misc/boinc.png

Seems to work - I'm not getting any work for Seti though...

[pfb]

Originally posted by FoBoT
mine ran out of work earlier this morning

i told it to cache 2-4 days of work, but it will only get 2 or 3 work thingys at a time, which is only about 1/2 hour of work

i think it is having a problem with the estimation of how long the work will take. since it thinks one work thingy is much longer than it really is, it won't download more at a time

i guess

I've set mine now to 3 to 14 days work - I get around 30-45 mins work now...maybe worth increasing that cache size to something similar...

[Helix_Von_Smelix]

I see no Free DC team has been created yet, and i can't download any work.

:Pokes:

[pfb]

There is a Free-DC (http://predictor.scripps.edu/team_display.php?teamid=15) team...and work seems to run out quite quickly for some reason...

I've found right-clicking on the Preditctor project in the project tab and selecting Update downloads work rather than waiting a random number of seconds (I find after 2 attempts that BOINC chooses a silly figure of 40 minutes to wait...)

[Digital Parasite]

Someone else was posting that the server right now isn't giving out any WUs to work on. Is anyone here actually seeing this or does it seem to be working well?

It should be interesting to see how this group does compared to DistributedFolding in the CASP6 challenge.

Since CASP6 is currently running, you would think these guys would have rolled out the client for testing before now.

Jeff.

[pfb]

I'm not getting any new WUs on clients set-up earlier but one I setup a few hours ago is working on already processed WUs...

/edit - getting new work now....

[Hua Luo Han]

funny : wif HT on, i get computation error, turn it off and its fine :scratch: :dunno:

[pfb]

similar if I run it on a dual processor PC - I think this version of BOINC has issues with running on more than 1 (logical or physical) processor :(

[excaliber]

There was an official post by one of the admins earlier. The WU shortage was them fixing something. Said there would be more up later (seems mine is back to processing). They are adding more targets, so they said that shouldn't happen again.

Processing time right now is only 10-ish minutes. They said that should increase, right now the processing is very minimal (structure only, some other stuff) Said, after stuff was worked out, that the WU's would be fleshed out and take longer.

[FoBoT]

i forced a retry and got a lot more work

[pfb]

might keep 1 PC on this just to see how the BOINC scoring system works (at the moment I feel it is unfair but just have to see)...shame the client doesn't work on multi-processor PCs...

[Hua Luo Han]

i tried running the linux client instead and it works fine on smp :D guess its the new windows client that needs some attention :rolleyes:

[pfb]

yep - got a reply from the organisers saying it's fine on Linux...offered to test any changes on the Windows side so we'll see...I'd prefer to leave a SMP box on it than a AMD64 box...

[excaliber]

Originally posted by pfb
might keep 1 PC on this just to see how the BOINC scoring system works (at the moment I feel it is unfair but just have to see)...shame the client doesn't work on multi-processor PCs...

Should work on dualies as well...theres options under the config xml file i believe

[pfb]

always fails on mine - the team said they'll have a look anyway...

[TeeJay]

I find it quite odd... that ~every workunit I have received credit for,
HaloJones has crunched the same workunit. Hmmm....:dunno:

[Digital Parasite]

Originally posted by TeeJay
I find it quite odd... that ~every workunit I have received credit for,
HaloJones has crunched the same workunit. Hmmm....:dunno:

Awwwww... You guys are like protein brothers or something.... :D

[HaloJones]

:cheers:


EDIT: How do you know, Teejay? Where is this information?

[ladypcer]

Hi guys. I'm new to the forum, but not to running DC projects.
I recently found out about the P@H project and I joined a team, but I don't think it's going to be an active team any time soon.

Are you guys going to be running an active P@H team?

I might be looking to switch.


LP (Lady Pc'er)

[Paratima]

Hi, Ladypcer. :hiya: Free-DC is a pretty convivial place. I'm not on that particular project, atm, so I'll let someone else talk about it. But in general, this is a great bunch of folks to crunch with.

[ladypcer]

I started way back with Seti screensaver when it first came out, then ran UD for about 1 1/2 yrs. More recently I've been running F@H and F@D. Predictor looks like it could be a good project to run when it really gets off the ground. I'd like to start out early, while it's in it's infancy.

[TeeJay]

Originally posted by HaloJones
:cheers:


EDIT: How do you know, Teejay? Where is this information?

Here is a link to user Ironbits... not IronBits... there is two...

http://predictor.scripps.edu/show_user.php?userid=108

If you click "VIEW" computers it will take you to Hostid 906...
Click on the Hostid 906 and it will take you to the summary page.
Scroll down to Results and click on that.

It takes you to all the results from that computer.
Click on the Workid and it will show who worked on that workunit.

You are 963... still my only verifier.
You are losing a few points with me being YOUR verifier...
Need to get a faster processor i guess... mine is xp32oo.

Crunch on Halo... nice job ! :cheers:

[FoBoT]

i still don't have any points :scratch:
how old will i be when i get some points? :geezer:

[ladypcer]

Ok guys, I don't know how it'll affect my points or your team's standings, but I just joined up with ya.

Now, I need to get some wu's...
I'm out of wu's on 2 machines.

:mad:

[Paratima]

LP: Very cool. Welcome aboard. Hmmm... not only that, you're quite near the top of the board. Very cool.

FoBoT: All this, and you want points, too? :p

[ladypcer]

Originally posted by Paratima
LP: Very cool. Welcome aboard. Hmmm... not only that, you're quite near the top of the board. Very cool.


Yeah, and I just downloaded and started running this a couple hours ago.
It's going like mad. :D

I have a fairly large "farm", and I have Predictor running on at least half of it, but if I can't get new work units... bleh..
I wonder why they are running out of wu's?

[Paratima]

I think the guy called FoBoT used 'em all up. :p

[ladypcer]

Originally posted by Paratima
I think the guy called FoBoT used 'em all up. :p

:D Hey FoBot! Give me some wu's! ;)

[ladypcer]

I just got some uploaded. maybe I need to go in and change how many days to queue up for..

[pfb]

just got some WUs after about 2 hours of getting the no work message - seems to be the T0198 WUs rather than the 196 or 197...

still got over 100 results to be validated though...from what was suggested on the forums at Prediction it may be that the units only get validated by a similar OS and CPU setup...

[ladypcer]

Originally posted by pfb
from what was suggested on the forums at Prediction it may be that the units only get validated by a similar OS and CPU setup...

Can you post a link to that, pbf?

[pfb]

it's BigJen's post (~halfway down) : http://predictor.scripps.edu/forum_thread.php?id=29

Homogeneous redundancy link : http://boinc.ssl.berkeley.edu/homogeneous_redundancy.php

not sure how true it is...

[FoBoT]

when it runs out of work, try poking it :Pokes:

if you right click on the project under the projects tab, you can manually retry the upload/download process. this has gotten me new work a couple times


i am not the work hog, i am only running it on my laptop right now :p

i think they have had a large influx of users this weekend, when i joined on friday the list of teams/users was much shorter


also keep in mind that this project is alpha/beta/testing stages, there will be bumps as they get it all sorted out

:drums:

[ladypcer]

Originally posted by FoBoT
when it runs out of work, try poking it :Pokes:

if you right click on the project under the projects tab, you can manually retry the upload/download process. this has gotten me new work a couple times


i am not the work hog, i am only running it on my laptop right now :p

i think they have had a large influx of users this weekend, when i joined on friday the list of teams/users was much shorter


also keep in mind that this project is alpha/beta/testing stages, there will be bumps as they get it all sorted out

:drums:

Yeah, I've been doing the manual upload/download.

True, it's still in it's infancy, so they are still working out the "why's and wherefore's".

I'm not sure I like getting the lower score based on your cpu and someone else's though.. hmmm..
I'm losing potential points here.

That could also pit users against other users.

[IronBits]

I noticed the client for predictor is OLDer than the one that SETI uses.

I installed the predictor client, ran it for a bit, joined SETI, that's when I found out the client was older, so I downloaded the newer SETI Boinc client.
Installed to a new directory (because I wasn't sure it was the same client)

Joined the SETI project, then tried to join the Predictor project, which worked.
So I stopped everything, removed the previous Predictor directory, renamed the BoicSETI dir to BOINC...

Linked to the Predictor project once again.

Seems SETI gets most of the work, even tho both are set for 50/50 processing time.
/me shrugs

Welcome ladypcer :cheers:

[HaloJones]

Originally posted by TeeJay
Here is a link to user Ironbits... not IronBits... there is two...

http://predictor.scripps.edu/show_user.php?userid=108

If you click "VIEW" computers it will take you to Hostid 906...
Click on the Hostid 906 and it will take you to the summary page.
Scroll down to Results and click on that.

It takes you to all the results from that computer.
Click on the Workid and it will show who worked on that workunit.

You are 963... still my only verifier.
You are losing a few points with me being YOUR verifier...
Need to get a faster processor i guess... mine is xp32oo.

Crunch on Halo... nice job ! :cheers:

They're going to need to be much clearer about this. I have a team user who has no points yet because no-one has validated his results. He's obviously unhappy because nothing is showing on the team list.

[TeeJay]

BOK... it looks like your duallie is having one heck of a time...

You have produced 430 results, but only 36 are good.
ALL the others have a computing error...

Just a heads up...

Crunch on...:thumbs:

>>TJ

[pfb]

The client doesn't run too well (or not at all) on duallies - they've suggested setting it to run 1 process rather than 2 as there are problems...

[PCZ]

When I finish a WU and upload it I want credit for it straight away.
Having to wait for others to return work before you get credit sucks.

Infact everything about Boinc sucks.

The future of DC and it can't even run on SMP machines :mad:

I will not now or at anytime in the future be contributing computing time to any project associated with Boinc.

[pfb]

It is possible to set BOINC to always allocate a set score rather than wait for validation - but this is up to the project...

The big thing about the validation scoring is that the lowest credit gets applied to everyone - why not an average of those who crunched the WU. Surely this means someone who got 10 for their work and someone else got 5 due to a faster setup, the person who got 10 wouldn't be too happy to lost half their points...

[ladypcer]

Well, for better or worse (P@H), I am putting the team higher up on the map.
I'll stick with this project and see where it goes.

[FoBoT]

i still don't have any points (Pending credit: 191.20)

what happens if another 1.06 Ghz pentium 3 laptop never comes along to "validate" my results?!? :cage: :gone: :hair:

i don't think i'll run it on my laptop anymore

maybe i'll do a pair of identical boxes at work. and since there is some issue about getting credit for the slow cpu, i'll use some nice slow celery 600 or pentium 3 733's

[pfb]

Originally posted by FoBoT
maybe i'll do a pair of idential boxes at work. and since there is some issue about getting credit for the slow cpu, i'll use some nice slow celery 600 or pentium 3 733's

from what I've seen, even if you are the only person running the same 2 PCs then neither will validate the other as part of the anti-cheating process...you could possibly get around this with 2 accounts (i.e. 2 different email accounts and account IDs) but as Predictor is in Alpha testing, the points situation may change...

(funnily enough no-one seems to be running an AMD64 3400+ with XP as none of those are getting validated - my AMD 64 3200+ with XP lappy is the only one validating :( - here's hoping that they release the fix for mulit-processor boxen under Windows soon)...

[Chinasaur]

ladypcer,

Welcome to FreeDC if this is your first project with us :hiya:

Apologies if its not your first and nobody said hello yet :)

Thanks for the on the project regardless :thumbs:

:cheers:

[ladypcer]

Originally posted by Chinasaur
ladypcer,

Welcome to FreeDC if this is your first project with us :hiya:

Apologies if its not your first and nobody said hello yet :)

Thanks for the on the project regardless :thumbs:

:cheers:

Thank you Chinasaur. :)

This is from the P@H forum-

Detailed News and Updates

6/14/04 - We have implemented a solution for a couple of problems that have emerged over the last days:
MFold was crashing on one cpu of a SMP and HT windows system. This was due to access to common files which failed for the second process. This has been fixed in version 3.06 of MFold. Oh, and don't forget to set the number of CPU's back to 2 in your preferences...
The progress bar for the computation on the Boinc client for Windows was not implemented yet. This has been done in version 3.06 of MFold.
The estimated time for work units will be more realistic (we hope). An average work unit on an average machine (let's say a 2 GHz P4) should take round about 2h. To achieve this we will kick out a lot of the current work units from the DB. This means that work units currently out on your systems will never be validated. Therefore the thing to do is to stop all Boinc clients you are running Predictor@home on. Delete all the P@H related directories and start again. Your accumulated approved credit will not change doing this, you are only making sure that you are not computing work units that are useless.
Other smalll things that have been fixed are the fact that you now can click on the top Predictor@home image to get to the home page, a bug that corrupted the url in the subscription emails has been fixed and the 'total/average credits' that were different on the top teams page should be more consistent now: the cache parameter has been set to 10.

[Thor]

Is that any good for an AMD K7 800Mhz with 512MB RAM and W2K?
anybody know?
Usually I would run DF on it , but the current protein is so slow that I switched that machine off....

Has anybody tested it yet on such a slow machine?

Greets Thor

[ladypcer]

Originally posted by Thor
Is that any good for an AMD K7 800Mhz with 512MB RAM and W2K?
anybody know?
Usually I would run DF on it , but the current protein is so slow that I switched that machine off....

Has anybody tested it yet on such a slow machine?

Greets Thor

The question to ask would be "would anyone else be running the same machine?"
The identical protein has to be run on an identical machine that belongs to another user, to validate your points.
That's a big sticking point.

[pfb]

Originally posted by ladypcer
The question to ask would be "would anyone else be running the same machine?"
The identical protein has to be run on an identical machine that belongs to another user, to validate your points.
That's a big sticking point.

they've relaxed that now - a couple of WUs I have are being validate by other CPU/OS combinations (my AMD64 3400+ has a XP 2500+ validating the WU)...

[Anteraan]

They had to. I really doubt the project could have survived with that validation model.

[ladypcer]

Maybe they haven't relaxed them enough. I have over 100 points waiting to be validated just today, and that's on a 2800AMD XP, XP pro, 512 ram.

[pfb]

Originally posted by ladypcer
Maybe they haven't relaxed them enough. I have over 100 points waiting to be validated just today, and that's on a 2800AMD XP, XP pro, 512 ram.


A looser homogeneous redundancy has been installed. The scheduler checks the kind of OS (Linux, Windows, or Darwin) and the kind of architecture vendor (AMD, Intel and Mac hosts) but does not check aspects like the edition or CPU speed.

That's what they're now using....so your WUs would be check for Windows and AMD - and also due to the number of people currently 'testing' the system there may still be a delay validating...

[ladypcer]

I have 3 out of 8 pending wu's that are being worked on by others. From a 2600 AMD to a 3000 AMD 64bit.

Maybe it'll even out overall. ;)

[Death]

Originally posted by PCZ
Infact everything about Boinc sucks.

I will not now or at anytime in the future be contributing computing time to any project associated with Boinc. [/B]

BWAHAHA!!!

That is my decision too.


PS. That was my idea for creating BOINC forums here =)))

[Petey]

I object to the points score for my Cele 2.6 being downgraded to be the same as the Pentium 800 used to validate my job.

Over 30% of credit lost. :swear:

[Digital Parasite]

Really? If your job gets verified by a slower machine, the points assigned to both people are those of the slowest machine?

But why would the points be less? Aren't the points assigned based on the protein you are checking? It might take the P800 longer to do, but in the end its the same results right?

Jeff.

[PCZ]

I just checked the Free-DC stats for predictor and Fobot has 0 points, nil ,nada.
Did you give up before returning any WU's ?

I was giving the project a second look to see if my initial impression was wrong.

[FoBoT]

it never gave me credit, no "granted" credit

see this page (http://predictor.scripps.edu/results.php?hostid=901)

i have 10 results that were never confirmed or whatever the process is supposed to be

i'll put it onto another box that has a more common cpu sometime and see if it will give me some points

[BrianM]

Hey people! Just joined Predictor.:drums: Hope to help a little.:cheers:

[Xelas]

Hi guys! Can this client be sneakernetted? I only have dialup access (33.6), and then for only a couple hours per day.

I have exactly nil, zero, nada experience with BOINC - I quit SETI back when a 166MMX OC'd to 266MHz was a red hot machine.:D

[BrianM]

Xelas,
Yes I believe it can if I understand you correctly.

[Xelas]

Yeah, great! Now, if only they enable new account creation . . . .

[BECL]

Hi all...

I would like to introduce myself... I'm one of the new names you see on the member list......

I also have a lot of first hand work in the source of BOINC (dev-alpha-beta-live) and Predictor(dev-alpha... hoping for beta shortly) and what is happening behind the scenes for anyone that wants to know...


I see the last comments was account creation, sneaker netting, etc....

I can answer those and others if you wish... just give a hollar....

Also, it may not be well documented, but you can run Predictor (or any BOINC) as a service.... I have all the info if desired...


For now, I'm going to be helping push up our numbers and if we want to do the 'beat the stats' game, there are a few simple things to do that will get a higher score and help counteract those who don't benchmark their computers (which is why some are so darn low).

Hollar / PM me if you have any questions..... just let me know how I can help.....


There is only one thing I CAN'T figure out... and that's how to keep Predictor from overrrunning 'Gridwin' on D2OL.... it shares everything else just fine... but really kills Gridwin (unless I baby sit it and make it run 'below normal' priority by hand). Can anyone help with that one?


Thanks,
Chuck

[nickth]

Two question.
Have they ironed out the bugs?
I saw that there was a new version 3.22 is it any better?

[BECL]

Originally posted by nickth
Two question.
Have they ironed out the bugs?
I saw that there was a new version 3.22 is it any better?


Nickth,

Most, if not all the app bugs are settled.... There are a couple basic issues still open:


1. the host itself is grossly overloaded because we are producing solid results and it's all being ready to be taken to CASP 6.... ( the competition about folding/predicting). So the server can be slow... and quorum is sometimes 2, sometimes 3.... this delays granting points..... but not a big deal... it is just hard on the host because it holds all our result records in shared memory (some 15,000 at a time) as WU's get cranked around


2. 3.22 is intended ONLY for those machines which are experiencing the 'Mfold app with Charmms data' problem and visa versa.... it contains more debug and will slow you down... it also may outright crash... it is a 'WTF DEBUG' patch.....

My recommendation is to keep an eye on things and if you get a batch of WUs
of the form xxxx_1_xxxxx_n_n_n (meaning gong around 3*n times).... DUMP IT.

(hate to say this.. but clear it all out and start fresh ..

1) shutdown executable
2) empty the project folder (down deep) of all project files (keep the executables)
3) empty the slots/0 folder (where the jobs run)


when you restart it will complain... hit 'reset project'... and you get a fresh load.....

the system will get the reset and reschedule and redispatch later... (it will also revalidate the WU to make sure the app and WU are in sync again).... This fixes your 'client_state.xml' file the easiest and safest way..


I have not had to do any resets since the bad data from the wu generator (2-3 weeks ago) got out of the system...... that is the one that also caused 'computational errors' for everyone.


3. Boinc 4.0 is about to come out and as soon as I can grab it... I will.... Dev/Alpha are ironing out issues... but everyone will love it..... No more MUTEXs... alll plain sockets..... SWEET!!!!! other issues got fixed too.... there are just a couple oddities for alll the different OS versions that are being tested and fixed as they are found... but it's about 95-99% ready to hit the street.... I will let you know when I have it and you can play if you want..





In summary of that long winded (yes, i am feeling better) dissertation... it's 500% more stable than it was.... The new servers will be here soon and when they are, the dual Itaniums get replaced with 2 - quad opterons..... Server side will be 100% able to keep up and account creation will definately reopen..... (I suspect it will reopen as soon as the first machine i ready).....

the only real issue now is generating work units... the server is max'd out! ... we have an error-rate now that is virtually non-existant..... (better than Seti/BOINC)... and WU's are created... dispatched, returned.... and that's it... BOOOM, DONE....like it shoudl be.


SO..... I say... Go for it.... I am.....all my machines are one it now.



PS: The 'total credits' for each user will be reset at some point soon... that just takes manual scripting to run through the DB.... and the boss doesn't want to touch anything with it being so stable and producing so much so fast....

Chuck

[BECL]

Folks,
I just did a check of the database.... from what I can see.. most WUs have a quorum set at 2. So the lower of the first two back win. Yes, unfair... but until we have a server than can handle the proper quorum of 3 (where the mean or median... forget which... is used) we are stuck with it as it is.....


However, the main point of this post..... if you set (in your general preferences) a min and max that differ by .05 (e.g. 1.00 days min and 1.05 days max), this will pretty much guarantee that Predictor updates your score every hour and you will receive a steady stream (or at least attempt to get it) of WUs..... ... since .05 days = 1.2 hours..... As a side point... with a steady stream 'history' of work produced... your RAC will decrease MUCH more slowly if you are shutdown due to lack of work for any amount of time.... 'bursting' is not advisable in Predictor.

You will have to build up a list of pending WUs... this is normal unless you are the 2nd or last CPU to report in. I find a typical 'backlog' of about 50-75 WU's depending on the processor... I have faster machines, so I am pushing closer to 100-120 pending WUs. This is normal......


If you are willing to dedicate a reasonably fast processor and have a stable connection.. then you can (and definately should) increase the disk write time to something up to the 600 second range. You don't get credit for writing a checkpoint file (50MB) to disk. BOINC 4.0 should fix that. So to maximize throughput, increase above the 30 second mark to 120, 240, 360..... whatever you are comfortable with.

Also don't be afraid to use the home, school, or work 'venues' for each different performance class of machine... this helps you BIGTIME in adjusting to each different processor class or make (intel vs amd... or HT vs non-HT, etc etc etc)

One last *edit* before ending this..... keep the GUI window closed... not minimized... actually "X" out of the GUI window.... it will run more like the CLI then and not poll the Mutex (which is a MAJOR slowdown).



My A'64 3200+ and FX-53 (both OC'd)... run with 2.0, 2.05 and 600 for reference.


Those two tweaks are two major differences that set Predictor and Seti apart.


Also, a long 'pending' list means you stand a good chance of 'setting the standard' for a particular WU.... no guarantee as someone could come along and undermine you, but it does help.

In all... the more steady, stable turn around time you can achieve, the better you will do with any BOINC project..... just remember each project has it's own quirks.

What I have given you here works best for predictor... giving the fastest "RAC" and Point climb possible.


PS: Since Predictor is multi-task... there is a way to get multiple host ID's and run 2 Predictors on single processor machines (which takes advantage of both hyperthreading and BIG advantage of AMD architecture)... I will see if I can find that info again and PM to those interested.



Sorry for rambling, just trying to help and share too much too fast.


That's the end of the edit's too...

Please comment, question, whatever.... I'm ready to contribute :)





Chuck

[BECL]

Folks,

Before I head out for a while on errands... anyone experiencing 'no schedulers responding' means we are hitting the server faster than it can respond..... BOINC 3.20 can handle this without problem.

The same goes for 'no work...'


If you wish manual intervention.... Open the GUI....

a) simply do a 'retry now' for individual (stuck) file transfers...... right click (retry)

or

b) project tab -> 'update now'.....


both send off the RPC to the server and push you back up the list.....

This is unfortunately common when someone comes in and 'dumps' a load of WUs
and then drains the WU queue..

Please be patient.... closing (X) the GUI during those idle times you wish to give it a rest.


General suggestion is to work up to a 2 day supply of work at the .05 delta min/max and you will be fine regardless of whatever delay BOINC tells you it is waiting on.

Once the surge is over, all your files will show up and queues will fill.
Same is true with uploads and 'ready to report'.... all will work out ... just let it run.

It's never failed me since I started back in May (BOINC)... and Predictor (June).


Hope this helps with peace of mind and avoids any frustrations..



Have a great day...
Catch you all in a little while.


Chuck

[Spore]

Predictor is a waste of time IMO. It isn't even Beta yet. Why spend time on it when you can risk losing stats - It's down now, again. I don't see why we should do it now in its current state.

[BECL]

An Update: 19-Aug-2004



Been running Predictor updated 3.20 client (from their web page)....
I wrote and found no need to run the 3.22 client.

I had 2 WUs with a problem; the pc spent time on them but still got credit for the effort.

I've run both the CLI (-win_service) and the GUI... both are solid.

It's been stable for 48+ hours: work unit, site, and (most important) points.

I've always been able to upload and to get work.

I've built up over 800+ points in pending credits (about 13 points at a time) and am reaching the 'balance' point. (mostly due to the lack of AMDs out there)

Our team RAC and Total Point scores are climbing.... and so are our standings.


The database work was done and it's been successfull w/ no damage to anything,
nor did it have any down time.


I've have been and am able to upload a completed WU consistently every 20-25
minutes and with quorum == 3, the scores are 100% fair...

Is anyone else out there "kicking the tires" at all?


I am also running D2OL and keeping up my part on that... it all seems to work together well. CPU balance is 49/51 or 50/50.


Please comment.


Thanks,
Chuck

[willy1]

Pretty hard to give it a try if you can't even create an account ..


:bang: :bang: :bang:

[BECL]

OOOPS... my fault.....

I will write to the admin and ask for the new accounts or see if I can pop
them in myself.....

How many want an account???? please PM me with your desired name and an email to put with your account..... I will get it done.


It is also possible to slide in under someone's account and then move your machine at any time...

Let me know.



Chuck

[PCZ]

Is anyone else out there "kicking the tires" at all?


I have put some boxes on it.
Are the stats likely to be reset?

[BECL]

No stats resets are planned... what is done is done.

Everything is good to go... nearly 72 hours now myself of solid running... only 4 times did I get a 'no schedulers responding, deferring for 1 minute'.... on the next try it was right there WITH work!!

I would say "Launch".

[PCZ]

Chuck

Are there any 3rd party stats pages ?

[em99010pepe]

Originally posted by PCZ
Chuck

Are there any 3rd party stats pages ?

Check these ones:

http://www.setisynergy.com/stats/index.php
http://www.boinc.dk/index.php?page=statistics

Carlos

[BECL]

Originally posted by PCZ
Chuck

Are there any 3rd party stats pages ?


Yes.... boinc.dk (the team) is providing stats..... stats are updated at least 1x a day, sometimes 2.


If you want the 'dirt'.... read on.....


You need to start with the main list and find your team #, and your user #, then fill in the following....



http://www.boinc.dk/index.php?page=user_statistics&project=pah&userid=YOUR_USER_NUMBER



and


http://www.boinc.dk/index.php?page=team_statistics&project=pah&teamid=THE_TEAM_NUMBER



I would fill in the values now,, but i (***embarassed***) realized I have not yet updated mine to the new values.... The best way to get your team # and user # are by starting at the top of boinc.dk and paying attention to the URL as you search and dig down... then just bookmark the right one.



I'll get the numbers, but I think you'll find them if you go to boinc.dk and select the different projects...ALL Boinc projects are now supported.


At this level, you can see all the details direct..... What Carlos shows is also perfectly valid.... depends on what you want.



Chuck

[willy1]

still no account creation possible --

Guess they don't need or want new users.

[PCZ]

Willy1

It is annoying that new users cant sign up.

Like a lot of projects they are using the Beta tag as a shield, they can excuse any errors because of it.
The FanBoys are jumping on everyone who complains, pointing out that the project is still in Beta.

It's about time they released a supportable client.
Climate Prediction goes public on the 26th is that going to be Beta ?

Boinc is not living up to the Hype.

Over the last 48 hrs I have built up pending credit of over 13,000 and this is rapidly increasing.
At the same time only 1500 points have been credited to my account.


PS

If you still want to participate IB has 2 accounts.
One of them is a variation on the spelling of his handle, big B little b.
Perhaps you could take over one of them if he still has the account ID.


Edit:

Things have improved quite a lot since I wrote this post, I am actually warming to the project.

New registrations are disabled until the end of CASP 6 :(

[willy1]

I really don't want to use someone else's account, nor do I want to pass my personal information on to someone I don't know ('BECL', alias 'Chuck' - who is this guy and how is he related to the project?) to set up an account for me.

I guess BOINC is just SOL. Seems like everything associated with Berkeley is FUBAR.

willy1

[em99010pepe]

If you want you can stay with my account. Just give me a valid email I will send to you the password.

[PCZ]

Willy1

If you want to Boinc, how about Climate Prediction (http://climateapps2.oucs.ox.ac.uk/cpdnboinc/)

It officially went public on the Boinc platform today.
It is a bit lonely being in a team of one :p

[willy1]

Climate Prediction was my first DC project - it's how I ended up here at Free-DC.

Last time I looked, I was still in first place on the team. Did they trash all the pre-BOINC credits?

[Welnic]

Originally posted by PCZ
Willy1

If you want to Boinc, how about Climate Prediction (http://climateapps2.oucs.ox.ac.uk/cpdnboinc/)

It officially went public on the Boinc platform today.
It is a bit lonely being in a team of one :p

Count me in. :cheers:

[willy1]

I'm crunching too.

SETI wouldn't give me any work, and Predictor doesn't want anybody.

CP gets the GHz.


willy1

[BrianM]

As of today we are in 9th place! Keep it up people!:elephant: :drums:

[em99010pepe]

From site:



Account Creation

The creation of new accounts is temporary closed. We will open again the creation of new accounts after the transition of our server to BOINC 4.x and the end of CASP (approximately the second week of September). Thanks for your patience.

[em99010pepe]

Originally posted by em99010pepe
If you want you can stay with my account. Just give me a valid email I will send to you the password.

I changed twice my email but account ID still remains so I can give my account. I'm sorry.

Carlos