Predictor Status ---- Updated as needed

[BC]

28-Jul-2004 10:00 UTC)
(posted 10:20 UTC aka... 3:20am PST)


Gentlemen (and the rest of you).

It is official.... I watched the transitioner / validator run
and it has confirmed for the 3rd time..


WE ARE IN SOLID 6th PLACE AND STILL GAINING..

My thanks to all .....

I will have some software update notices for you all shortly (working on code fixes)

We will be able to run faster and bigger jobs (HOPEFULLY) in the next rev of both MFOLD and also have a faster, fully stable Charmms...

It's all being done under Cygwin for the windows platforms...

This also will allow creation of the proper MAC versions as well.

64 bit versions are on the slate and at Pri #2... Pri #1 is getting it solid of course.


PS: Let's get that Team Logo and Score php code in place... We are going to need it.!!! LOL


BC.

[BC]

Guys,

There are a whole BUNCH of the easy, quick credit WUs to be grabbed.


GET THEM FAST BEFORE THEY ARE GONE!!!!


I got all I could get...


BC

[BC]

Guys,

There are a whole BUNCH of the easy, quick credit WUs to be grabbed.


GET THEM FAST BEFORE THEY ARE GONE!!!!


I got all I could get...


BC

[Anonymous]

Thanks for the tips BC i grabbed a few and edited my profile so i could have a longer supply on my HD, i also set up another machine but its not as gd as this one (3200) so i cant double my WU's but it all helps lol

[Empty_5oul]

btw that was me, soz i though it logged in but obviously something didnt like it and i ended up annonymous :(

[Anonymous]

btw that was me, soz i though it logged in but obviously something didnt like it and i ended up annonymous :(


Every bit helps.... I was working on Mfold at the time and missed the DB re-release...

I now need to do what's right and come up with an AMD compiler optimized version of MFold.... :)

I find it interesting that all the leading teams have BOINC alpha team members on them.....

Now we are a leading team and gaining ground fast, huh?


As for grabbing WU's..... grab all you can handle (everyone). pass the word... and just delete the T0235 WU files from your project file... the DB will skip right over it.... it has to be that way until I fix Mfold...

Now, as for creating an AMD optimizzed version.... what's the best way???? who knows GCC REALLY well???? I don't think the intell instruction scheduling rules will help us, but it sure helps the Pentium teams.

[BC]

that previous post was me.... forgot to sign in.

since my post last night.... we have moved up 100+ RAC and a LOT of points. A lot of other teams have as well. The database is getting cleared and 'pending credits' are getting cleaned up fast.

The jobs being re-run are known good and should not cause problems.
I've scanned the main DB. Of the 5000+ WUs released, only systems running older versions of software (OS's) or version 3.10 of MFold have had problems. Mfold 3.11 should download clean automatically and run clean.

some reports i have of errors have been linked to systems not having all of the released MS updates (Pentium, Mac, and AMD on windows. Linux users need to check their respective libraries). There *might* be an error or two in the data, but so far no pattern emerges other than machines that are not 100% up to date from Microsoft.

If you get lots of errors, please check your machine and report to me if updating it does not correct the problem. One or two errors are ok, but I would like a PM about them as well if you don't mind.... just give me the Host ID, WU or ResultID, and formal "T" job name (and your userid of course).

To date, only T0235 is known to blow the memory limits (it's Fortran) and what I am fixing.

When T0235 and the next block do not blow memory, as would it's Charmms counterpart since they share the same libraries, We will be in Beta mode.

Note: I'm working on a fix for the runtime priority of 'low' when it should be 'below normal' .... At the least, I will show you all how to set the value to 'below normal' to improve system throughput and pull Predictor up out of the OS noise.


Thanks,
BC

[nickth]

BC
In my pending list i,ve quite a few jobs like this.

415968 6455 11 Jul 2004 14:56:54 UTC 13 Jul 2004 18:50:14 UTC Over Success Done 4036.015625 14.80 0.00
443890 2190 12 Jul 2004 9:12:54 UTC 13 Jul 2004 6:46:46 UTC Over Success Done 3757.893586 16.49 0.00
415967 14536 10 Jul 2004 22:57:20 UTC 10 Jul 2004 23:12:09 UTC Over Success Done 0 0.00 0.00
415966 9104 11 Jul 2004 5:34:14 UTC 11 Jul 2004 15:23:08 UTC Over Client error Downloaded 0 0.00 -!-

the bold client is mine.

[Empty_5oul]

i take the mfold dl's automatically, some of my jobs atm r 3.10 and some 3.11 over time they will all become 3.11 and when next updated 3.12 as you (the writers) release another one.

With the priority is it as simple as opening task manager right clicking and making it what you want ??
as ur a writer ur probably the person to ask - I was wondering if it could be made possible for the user to decide how much maximum percent of their processor was used on predictor. I would definatley find it helpful cause predictor could be runnign more while i'm on the computer trying to do other things - when i ran AVG and predictor both set to "high" priority the sytem didnt like it (started crawling and eventually crashed) as they both normally use 100% processor.

Anyway im going on holiday tomorrow so for a week i wont be adding to the average, but when im back im gonna build a new machine and have predictor on that 24/7 as well as 90% of the day on this (3200+).

[BC]

BC
In my pending list i,ve quite a few jobs like this.

415968 6455 11 Jul 2004 14:56:54 UTC 13 Jul 2004 18:50:14 UTC Over Success Done 4036.015625 14.80 0.00
443890 2190 12 Jul 2004 9:12:54 UTC 13 Jul 2004 6:46:46 UTC Over Success Done 3757.893586 16.49 0.00
415967 14536 10 Jul 2004 22:57:20 UTC 10 Jul 2004 23:12:09 UTC Over Success Done 0 0.00 0.00
415966 9104 11 Jul 2004 5:34:14 UTC 11 Jul 2004 15:23:08 UTC Over Client error Downloaded 0 0.00 -!-

the bold client is mine.

Those WUs , stuck at zero are being worked on... we all have them unfortunately.. the transitioner & validator (which grants credits) never ran.... those will be reset to the state 'validator_state_init'.... then they will be validated and credit granted. they are not lost... but that will clear up whats left of pending. that is one flaw in the boinc engine.... the author will help fix that himself.

what you are doing now should be working ok and granting ok as each WU completed.

BC

[BC]

i take the mfold dl's automatically, some of my jobs atm r 3.10 and some 3.11 over time they will all become 3.11 and when next updated 3.12 as you (the writers) release another one.

With the priority is it as simple as opening task manager right clicking and making it what you want ??
as ur a writer ur probably the person to ask - I was wondering if it could be made possible for the user to decide how much maximum percent of their processor was used on predictor. I would definatley find it helpful cause predictor could be runnign more while i'm on the computer trying to do other things - when i ran AVG and predictor both set to "high" priority the sytem didnt like it (started crawling and eventually crashed) as they both normally use 100% processor.

Anyway im going on holiday tomorrow so for a week i wont be adding to the average, but when im back im gonna build a new machine and have predictor on that 24/7 as well as 90% of the day on this (3200+).


you are right, setting the priority too high prevents normal services from running and the system will crash.

for anyone who has time.... using task manager to bump up the priority to 'below normal' by hand gives the best results.

'below normal' was decided to be the priority to run at because it's up out of the 'idle' class (idle class::low priority) .... and put it (normal class:: below normal priority)... the numeric priority it should be at is 6 or 7... it is at 2 or 3 right now... depends on OS and revision level.

explorer, ie, and all other services run at (normal::normal) aka 8. you can run predictor jobs at 8 but it will slug a bit... just be carefull and not ask too much of the pc when doing such things.

[Beerknurd]

Mine are set to Below Normal. If I set them to normal it makes other applications run slower. Below Normal is the best. I agree with you BC.

[Beerknurd]

If you haven't already.....

Please Cast Your Vote!!!!! (http://amdusers.com/modules.php?name=Forums&file=viewtopic&t=542#4535)

[BC]

1. The mathematics wizards have been busy working on new Charmms code and a few errors are getting distributed... Please dont be alarmed at
failures or Charms/Mfold mis-matches. A few bugs to go, but please expect progress...


2. Per the home page.... Mfold energy levels adjusted... Now to complete the code optimization.


3. Here's the info on how to get our team stats and also to get your individual user stats and identical performance curves.

4. Expect some final tweaks and adjustments to code for Charmms and Mfold.... we're getting down into the tight stuff.... and it's almost there.


5. We as a team are solid 6th ... nobody even close... Keeep up the great work..... 5th place is within reach !!!!!!




(This is the XML / PHP code I would like to see us run for ALL our DC projects) a REAL stats page.)



------------------ Our stats ------------------------

Direct:
http://www.boinc.dk/index.php?page=team_statistics&project=pah&teamid=17

Code:

http://www.boinc.dk/index.php?page=team_statistics&project=pah&teamid=17

BC

[Ototero]

Chuck,

My brain cell is at it again...


What if.....

Underclock computer, get predictor to benchmark, then back to overclock.


Does that work ???

http://boinc.mundayweb.com/predictor/stats.php?userID=43

[Beerknurd]

hmmmm... good point... Chuck??????

[AMDave]

Chuck,

I see the symptoms of problems with Charmm only on the Celewrong -2000. The other CPU's are Ok. (PIII-800 + AMD Dur-1300)

What gets me is that I continually got more WUs on the Celerwrong than it could possibly handle. Now It's getting even more and *every* Charmm WU on that PC still says "Downloading" even though I watched them all download. The Charmm Wus on the other 2 CPUs are all fine.

I will watch and see if the Wus get processed but I am guessing they won't. BOINC has never quite behaved properly on that CPU. The Cel's are all crap these days. my PIII-800 and Duron-1300 beat the pants off them even though the Cel/ is more than 2 x the FSB speed of the PIII. I see this on all of the PCs we have at work processing data - they just chunk.

Have upgrade plans - will drop Intel...

--Dave--

[BC]

Chuck,

My brain cell is at it again...


What if.....

Underclock computer, get predictor to benchmark, then back to overclock.


Does that work ???

http://boinc.mundayweb.com/predictor/stats.php?userID=43


Stu,
If you do that,,, the 'cpu time' required to complete a job will be less than what it expected... you will get less credit. HOWEVER, there is a point in that bell curve where you benefit, but I have not found it.

I have found the best way to benchmark is shutdown EVERYTHING... including internet access services... the works... ALSO, put BOINC at HIGH PRIORITY..... then benchmark.

After doing the benchmark..... Exit boinc.

Reboot and restore normal ops.

Now restart Boinc, update and what you get is normal ops with 'higher than normal' benchmarks and expectations...

you get more points that way because it's taking you LONGER than it expected to do the work.... the difference is a LOT! This works because the ratio between your benchmark and the theoretical computer establishes a number which is now HIGHER than what you actually produce. Therefore WU_Base * (Actual CPU / Expected_CPU_per_WU)
gives the elevated 'Claimed credit' above the Base_expectation for each WU. You may still lose those extra points, but at least you tried. NOW, if a slower CPU does come in..... and claims more.. you both win.

Hope that math wasn't too bad.

[BC]

Gang,
There are a few faulty WUs bouncing around the grid. The code runs fine, but it fails to write and send the results correctly because of info in the WU instructions.

If you see ANY Charms WorkUnit of the form:

name_1_number_digit_digit ...... kill it and move on.

If any are in your projects folder fitting that description... DELETE THEM IF THEY ARE 3.10, 3.11 or 3.12

BC

[BC]

Chuck,

I see the symptoms of problems with Charmm only on the Celewrong -2000. The other CPU's are Ok. (PIII-800 + AMD Dur-1300)

What gets me is that I continually got more WUs on the Celerwrong than it could possibly handle. Now It's getting even more and *every* Charmm WU on that PC still says "Downloading" even though I watched them all download. The Charmm Wus on the other 2 CPUs are all fine.

I will watch and see if the Wus get processed but I am guessing they won't. BOINC has never quite behaved properly on that CPU. The Cel's are all crap these days. my PIII-800 and Duron-1300 beat the pants off them even though the Cel/ is more than 2 x the FSB speed of the PIII. I see this on all of the PCs we have at work processing data - they just chunk.

Have upgrade plans - will drop Intel...

--Dave--

Dave,

Per my post above this..... DELETE ALL CHARMS until it's fixed and move on. Charmms is from Harvard, not local. I am not allowed to even see certain sections... and the source of the problem.

*EDIT*

IT appears they figured it out on their own... which is GOOD NEWS... so *FINALLY* it's cool, but *DO* be carefull of the 3.11 and 3.12 Charmms jobs. They are a waste of CPU time. You were right, as was I...

Thank you VERY VERY much for your input... It was passed on, but not acknowledged via return email.

*END*


Let's take 5th place on Predictor, shall we??? Once 5th is secure... we can rebalance and I can put more horsepower up on D2OL.

As for dealing with the apparant corrupt install on your one machine.... let me know if you want help... but a detach... exit... overwriting install of 3.20 boinc pure... and re-attach will fix it.... You will get a new host ID... so remember which CPU to merge (get the date & time or whatever you need to know which is which) and then let the database handle it from there.

As for upgrading.. you may be one of the first end-users to get one... but the Sempron is here.... I submitted the article to the Admins for posting as well as posted it under hardware.




PS: My replacement processor (I had a failure) is on the way... I hope to have it running before I leave Cali, but probably will have to wait another week until I get home (it's a 5 day trip with load, drive and unload/setup)

[Ototero]

Thanks Chuck,

I knew there had to be some way to "adjust" the credits.

I will do it tonight when I get home. Also I'll check the errant charm units.


Stu

[Beerknurd]

I didn't get the charm message till I was running the last 2... :-(

[Ototero]

http://www.boinc.dk/auto.php?user=2879&project=pah&input=1088692908+-+Moonman&layout=1088692908+-+Moonman.JPG

[BC]

Question: Team "What's in your box" seems to be made up
of well over 50%... more like 75%-80% AMD processors

Do they meet our membership criteria?

If so, would someone care to invite them in?
It would boost them AND us.

(I'm thinking of how to take our team and boost us and the other team; still keep the AMD to intel ratio as high as it is; and grow into an even more dominant force on the DC circuit.

Would someone please comment and/or care to do some writing?

BC.

[em99010pepe]

If so, would someone care to invite them in?
It would boost them AND us.

BC.

Which guys?

It's funny but I have the same goal as you.

I asked these guys (http://www.mysterybyte.com/) if they wanted to be an AMD Users sub-team, just like Dutch Power Cow Team. The answer was negative.
If you don't know DPC is formed by sub-teams. That's why they have such power.

[nickth]

looks like predictor has crahed again. :roll:

[chaz]

Question: Team "What's in your box" seems to be made up
of well over 50%... more like 75%-80% AMD processors

Do they meet our membership criteria?

If so, would someone care to invite them in?
It would boost them AND us.

(I'm thinking of how to take our team and boost us and the other team; still keep the AMD to intel ratio as high as it is; and grow into an even more dominant force on the DC circuit.

Would someone please comment and/or care to do some writing?

BC.

I think it's a good idea, however, a poll or vote to get the input and/or opinions from the rest of the team should be in order, and if in agreement, contact be made by one of our fearless leaders..

[Beerknurd]

Question: Team "What's in your box" seems to be made up
of well over 50%... more like 75%-80% AMD processors

Do they meet our membership criteria?



I don't meet the membership criteria....... :-(

[nickth]

Gang,
There are a few faulty WUs bouncing around the grid. The code runs fine, but it fails to write and send the results correctly because of info in the WU instructions.

If you see ANY Charms WorkUnit of the form:

name_1_number_digit_digit ...... kill it and move on.

If any are in your projects folder fitting that description... DELETE THEM IF THEY ARE 3.10, 3.11 or 3.12

BC

A few work WU's that are faulty thats a bit of an understatment i lost a whole nights work on both computers to these $#@#^&&^$ charms and as for deletting them they just re load back up and thats a lot of them if you have setting 1.00 to 1.05. So i am going to be not running predictor until this problem is fixed because i just car'nt sit here and babysit it until its fix.

[Beerknurd]

So none of the charm WU's that you process get counted...???? Please respond. I have alot of them too. If I delete them they just come back... What do I do????

[nickth]

So none of the charm WU's that you process get counted...???? Please respond. I have alot of them too. If I delete them they just come back... What do I do????

No you will not get credit for them as they will return an error -1. recomedation shut down client untill problem is fixed. they say in the forums that will be this evening PST.

[Beerknurd]

that sucks...... Predictor is very unstable. I like running it, but there seems to be a,lot of problems.

[BC]

that sucks...... Predictor is very unstable. I like running it, but there seems to be a,lot of problems.

Beerk,
did you vote, and would you mind reposting your comments in the poll?

BC

[Beerknurd]

I posted my thoughts on the poll thread. :cool: