60% of my results are not granted credit
21 Mar 2005 16:01:14 UTC
Topic 188653
(moderation:
Hi i am running the Linux version, and over 60 % of my results are not being granted credit, What gives with this code ? Heavy frustration with this code, I'm about to give up on EAH
Robert Somerville
60% of my results are not granted credit
)
> Hi i am running the Linux version, and over 60 % of my results are not being
> granted credit, What gives with this code ? Heavy frustration with this code,
> I'm about to give up on EAH
There are a couple threads about this with at least a little info in them.
http://einsteinathome.org/node/188052
http://einsteinathome.org/node/188420
Last official word on it that I've seen was from Bruce on March 14, when he said: "Teviet Creighton is currently doing some work on the validator and studying some of these results. He'll respond to this thread when he's learned a bit more."
And that's where we still stand...waiting on some word from Teviet Creighton.
Darren
It's a very strange thing. A
)
It's a very strange thing. A lot of people running the Linux version is getting invalid results, with no granted credit...but not me
I'm a newbie, I joined EAH twelve days ago, and my Linux machine has generated 11 results with granted credit. I have also 3 completed results still pending for granted credit, but no one invalid, or with zero credit.
Now I have looked at Robert's results and the invalid ones show a lot of "stderr out" messages, like "resuming computation at..".
This doesn't happen in my own results.
Are you stopping an restarting your BOINC jobs very often?. Or are you runnig several BOINC projects?. My machine is an e-mail server that runs continuosly, I never stop it. And I am participating in EAH only.
Well, I'm just wondering about what's happennig, I have no response. And waiting on some word from the developers, too.
occasionaly the the one
)
occasionaly the the one machine is rebooted to Windoze ~every 3 days. the resuming computation is the result of somebody using the machine, and the Boinc/EAH going inactive as per my preferrences. The other machine (2 CPUs) is always in Linux & never rebooted.
What version of linux are you using ???
Robert Somerville
I'm running linux (debian
)
I'm running linux (debian testing) and have gotten almost 20 results, so far none invalid. Several are still pending though. My other windows computer has however had a few invalid results.
I use SuSE Linux 9.2, kernel
)
I use SuSE Linux 9.2, kernel 2.6.8-24.11-default
I have 1 gentoo 2.4.20 and
)
I have 1 gentoo 2.4.20 and the rest are mandrake (either 2.4.19 or 2.6.3).
Initially, I was getting a very large percentage of errors across all 5 systems. Since the modifications that they've already made to the validator, almost all of my invalid results are either on gentoo 2.4.20 or mandrake 2.6.3 - and the number of invalid results on those systems has dropped considerably (from about 70-80% to about 30%).
> occasionaly the the one
)
> occasionaly the the one machine is rebooted to Windoze ~every 3 days. the
> resuming computation is the result of somebody using the machine, and the
> Boinc/EAH going inactive as per my preferrences. The other machine (2 CPUs) is
> always in Linux & never rebooted.
>
> What version of linux are you using ???
The Linux version does not matter.
I run it with some LD_LIBRARY_PATH trick on Suse 7.3, without the trick on suse 8.1 and 9.1 and also on Mandrake 10.
I think it is more a problem with stoping/restarting the client. I cannot proof that, but I have some feelings that these errors are more likely in that case.
> I think it is more a
)
> I think it is more a problem with stoping/restarting the client. I cannot
> proof that, but I have some feelings that these errors are more likely in that
> case.
The problem is related to the way the different OSs do their math. In the bulk of the invalid results in question, the result was sent to 2 windows systems and 2 linux systems. Both OSs are returning matching results to each other, but different from the other OS. Since 3 of the 4 results have to come in to establish the canonical result, which ever OS gets 2 in first is the one that gets the credit. There are some instances where linux is getting the credit and windows isn't - if the linux systems both get their results in first - this happening is just the exception rather than the norm (maybe partly because the einstein linux executable is so much slower than the windows executable in actually doing the work).
They have adjusted the validator to try to compensate for the differences (with some obvious improvements), but they just don't seem to have gotten it quite right yet. Presumably, they are still working on it.
> The problem is related to
)
> The problem is related to the way the different OSs do their math. In the
Replace OS by compiler and I fully agree ;^)
The validator seems to be
)
The validator seems to be accepting all my results, now.
You can look at the last WU I have returned:
http://einsteinathome.org/workunit/584451[/url]
I have gotten granted credit, despite the fact my Linux machine was the last that reported results. And the first three were Windows machines.
I have completed 16 WU's up to now. 3 of them are still pending for credit, but the other 13 are valid, with granted credit. No invalid results by the moment:
http://einsteinathome.org/host/70869/tasks[/url]