Hi, I have noticed that several of the work units I have pending are going to error out and lose seti data, as i have a machine that has not had an error before, but when it runs against a CUDA GPU, they do not match,as in this case, http://setiathome.berkeley.edu/workunit.php?wuid=677997049 which I think will error out and be tossed out....but in this case, same machine, the third run of the WU is being done by a cpu, http://setiathome.berkeley.edu/workunit.php?wuid=677765004 and i think should match mine machine....

anyways, just thoughts on how we are crunching...

Nick

---

Yup. I have seen that also (because my RAC reduced each day) on all of my Rigs since last outage.
Several Task waiting because the Status is "Completed, validation inconclusive" with 2, 3 or 4 Results with different Client Software...

Helli

---

A loooong time ago: First Credits after SETI@home Restart

Hi, I have noticed that several of the work units I have pending are going to error out and lose seti data, as i have a machine that has not had an error before, but when it runs against a CUDA GPU, they do not match,as in this case, http://setiathome.berkeley.edu/workunit.php?wuid=677997049 which I think will error out and be tossed out....but in this case, same machine, the third run of the WU is being done by a cpu, http://setiathome.berkeley.edu/workunit.php?wuid=677765004 and i think should match mine machine....

anyways, just thoughts on how we are crunching...

Nick

http://setiathome.berkeley.edu/workunit.php?wuid=677997049
http://setiathome.berkeley.edu/workunit.php?wuid=677765004

Looks like one idiot running old V12 opimized on a Fermi and 6.09 got false -9 overflows. Would need to go to CPU or maybe somebody running x32f which produces much less false -9.

I've no idea if the correct result will be thrown out when it clashes with two -9's.

Bad luck - the errors are not on your side.

Edit: correction. two idiots with V12 on a Fermi

---

Carola
-------
I'm multilingual - I can misunderstand people in several languages!

Edit: correction. two idiots with V12 on a Fermi

Both hosts (5305178, 5257703) are known offenders already on Joe Segur's list.

so i just happened to get matched with a pair for miss configured GPUS, and not all, or even most do this?

Nick

---

Looks like one idiot running old V12 opimized on a Fermi ...

Too much money - too little brains ;)

Statistics like this make me laugh - or cry. Not decided yet :/

Number of tasks completed 3665
Max tasks per day 100
Number of tasks today 224
Consecutive valid tasks 0

(Possibly this has been discussed a million times, but why is "Max tasks per day" not reduced to ZERO for such hosts?)

---

Petition against 1366x768 glare displays: http://www.facebook.com/home.php?sk=group_153240404724993

(Possibly this has been discussed a million times, but why is "Max tasks per day" not reduced to ZERO for such hosts?)

Yes it was discussed.
Looks like FARMI not compatible V12 binary can correctly process tasks on some ARs. This allows small amount of validations. So, quota system is ineffective to inhibit such mis-configured hosts.

so i just happened to get matched with a pair for miss configured GPUS, and not all, or even most do this?

Nick

Exactly so. The tasks affected will be re-checked by another computer, and in the vast majority of cases, the result from the mis-configured host will be thrown out.

Unfortunately, the bad hosts get through (and waste) a huge number of tasks every day. 5257703 has downloaded about 2,000 tasks (over 700 MB of data) in the last 24 hours. So you come across their droppings more often than you might expect.

Given that they are consuming a non-trivial proportion of a scarce resource (bandwidth), I wonder whether the time has come to suggest that the hosts Joe identified should be blacklisted by the project (until/unless the software installation is corrected, of course).

We need a Server based Blacklist. LOL

hehe

Helli

---

A loooong time ago: First Credits after SETI@home Restart

Given that they are consuming a non-trivial proportion of a scarce resource (bandwidth), I wonder whether the time has come to suggest that the hosts Joe identified should be blacklisted by the project (until/unless the software installation is corrected, of course).

Or to improve BOINC's quota system. It should protect from such cases too....

Looks like one idiot running old V12 opimized on a Fermi ...

Too much money - too little brains ;)

Statistics like this make me laugh - or cry. Not decided yet :/

Number of tasks completed 3665
Max tasks per day 100
Number of tasks today 224
Consecutive valid tasks 0

The problem with looking at "Consecutive valid tasks" is that anyone who has an invalid task for whatever reason will have it set back to zero, a more accurate assesment would be to look at the percentage of valid tasks over a set time period or specified quantity of work units, unfortunately this would probably be too server intensive.


Kevin

---

Kevin

Looks like one idiot running old V12 opimized on a Fermi ...

Too much money - too little brains ;)

Statistics like this make me laugh - or cry. Not decided yet :/

Number of tasks completed 3665
Max tasks per day 100
Number of tasks today 224
Consecutive valid tasks 0

The problem with looking at "Consecutive valid tasks" is that anyone who has an invalid task for whatever reason will have it set back to zero, a more accurate assesment would be to look at the percentage of valid tasks over a set time period or specified quantity of work units, unfortunately this would probably be too server intensive.


Kevin

Not too server intancive. to calculate such proportion only number of failed tasks + number of total tasks are needed. Number of tasks today already kept. Nothing prevent to keep number of invalid tasks today too.
The big problem with "daily" based % - task validation can be much later than task reception.
So better to keep "total" % for host.

Call me an extremist but...
Why not have: Consecutive valid tasks 0 = quota 1/day ?

If you get an occasional invalid it will build back up fast..
also It will encourage running with an eye to accuracy.

Just a thought...be gentle.. ;) *ducks*

---

"The most exciting phrase to hear in science, the one that heralds new discoveries, is not Eureka! (I found it!) but rather, 'hmm... that's funny...'" -- Isaac Asimov

We need a Server based Blacklist. LOL

hehe

Helli

The facility exists: http://boinc.berkeley.edu/trac/wiki/BlackList (though there is some suggestion it may have been broken by the app_version breakdown). We'll never know unless the project decide it's enough of a problem to intervene.

Call me an extremist but...
Why not have: Consecutive valid tasks 0 = quota 1/day ?

If you get an occasional invalid it will build back up fast..
also It will encourage running with an eye to accuracy.

Just a thought...be gentle.. ;) *ducks*

well, it willharm project performance but will not solve problem.
Occasional valids will build up quota fast too.
Something that takes host history into account is needed...

EDIT: history is total % of validation for host for example. Hosts with low % should be penalized for long time aven if last few results are good ones.

Not too server intancive. to calculate such proportion only number of failed tasks + number of total tasks are needed. Number of tasks today already kept. Nothing prevent to keep number of invalid tasks today too.
The big problem with "daily" based % - task validation can be much later than task reception.
So better to keep "total" % for host.

+1

---

Petition against 1366x768 glare displays: http://www.facebook.com/home.php?sk=group_153240404724993

I've got one here. Because my set up is new I keep checking for invalids. Today I found this one:

http://setiathome.berkeley.edu/workunit.php?wuid=677758481

The trouble is if you look at the record of the two other 'valid' computers they don't look so good. Who would you trust? My Mum used to tell me two wrongs don't make a right.
Not saying my machine is right but it doesn't look as wrong as the others.

I've got one here. Because my set up is new I keep checking for invalids. Today I found this one:

http://setiathome.berkeley.edu/workunit.php?wuid=677758481

The trouble is if you look at the record of the two other 'valid' computers they don't look so good. Who would you trust? My Mum used to tell me two wrongs don't make a right.
Not saying my machine is right but it doesn't look as wrong as the others.

made link clickable.

Yours is the right result that sadly (and fatally for the science database) was thrown out because it paired with two black sheep running a faulty app (or rather an app not usable on Fermi).

bottom line - we are still getting false results inserted into the database because of Fermis set up the wrong way.

---

Carola
-------
I'm multilingual - I can misunderstand people in several languages!

I have an inconclusive . I have 30 spikes, he has 31 pulses. Its from my CPU his fermi with V12 apps. Both -9 overflows. Wonder who will validate it? Its our buddy 5472266. here is the linkhttp://setiathome.berkeley.edu/workunit.php?wuid=679452515

---

[/quote]

Old James

I have an inconclusive . I have 30 spikes, he has 31 pulses. Its from my CPU his fermi with V12 apps. Both -9 overflows. Wonder who will validate it? Its our buddy 5472266. here is the link http://setiathome.berkeley.edu/workunit.php?wuid=679452515

Stock Linux CPU app with a 1.9 day turnround - plenty of time to place your bets while we wait. I'll go for a second inconclusive and another resend ;-)