Alien88,
Now that we cleared out everything except for first and second pass, can you briefly comment on the logic of the current setup. All first pass users only get first pass and second pass only get second pass wu or split in ratio? Thanks.![]()
I'm not sure I've ever seen such a pristine set of queues since I started doing SoB all those years ago!
Well done everyone, and thank you.
Now to kill some more k!
Thanks, I did read that explanation a while back. I just wanted to make sure it is still the same.
Like Vato said, we've been thru quite a few hoops lately it is nice it has stabilized. Too bad DPC's stampede seems to be slowing down. Hope the next big guns come in blazing away. Let's get us a Prime.
Only a few dropped-tests showing up every now and again. Just over 5k tests to empty first-pass - runtime per test is significantly higher now - I'd guess 4-6 weeks before first-pass will run out.
Ya but have you looked at the currently pending test, 14399 with 957 active users in last two weeks. Those are a lot of pending test. Have they been dropped and put back in circulation? Are we waiting for someone to submit a result. The 90day stats are continually getting bigger. http://www.seventeenorbust.com/stats/oldTests.mhtml
C'est la vie.
At least it got noticed - thanks for fixing so quickly.
dropped-tests seems to be oscillating between 5.1-5.3k - not ever getting lower for the last few days.
Not looking nice for progress.
Is this lots of very old tests that are now being expired correctly, or lots of new tests being dropped? (for reasons unknown, but possibly including: DPC Stampede is over, folks don't like the CEMs/s reduction on these smaller n values, random bad luck)
Probably some combination of all of those things. Plus, lots of people haven't fully switched over to dropped-tests yet. I know out of my 15 active tests, only 2 are dropped-tests, the rest are first-pass, and even still a few second-pass being crunched on by my dual-p3 machine. Once we all get fully switched over, I think it should pick up.
I'm not sure why dropped-tests would be acting like that. There aren't all that many old tests, so it'd have to be manual expires.
enderak: the server will hand out dropped-test as a first priority, if the n value is less than whatever is in first-pass. So for most of these tests, people don't need to switch over, they're getting them automatically.
Alien88: Sorry - by switching over, I meant finishing up first- and second-pass tests and automatically switching to the dropped tests by way of requesting a new test.
It's only been 5 days since the dropped-test queue was filled, so if it takes (on average) 10 days to complete a first pass test, then only about 1/2 of computers would have switched over by now.
And at last, the dropped-tests queue has been emptied.
My new guess is that we'll need first-pass repopulating in about 5 weeks-ish.
Doomi 33661•2^14078808+1 80.238.147.59 Sep 16 2007 May 28 19:13 48 % 61006
this one seems completely useless
like this one
jkusuda 33661•2^13826592+1 67.112.246.235 Aug 4 2007 Jun 5 13:49 33 % 33972
so maybe you can write a mail to users that currently crunch tests with primes already find and ask them to switch to sieve? ))
Useless, yes. Worth the hassle, no. These slow computers are not a big loss to the project. If the user runs tests without paying attention to what he is testing, it's none of our business, I think, as long as he is not obstructing the rest of the project. H.
___________________________________________________________________
Sievers of all projects unite! You have nothing to lose but some PRP-residues.
It's a catch 22 situation.
I don't think we want to end up with a project where you can only participate with a very fast modern cpu that's crunching 24/7/365 - but some of these are clearly a bit too far in the other direction. A couple of them look like they're running on a 286 thats only turned on for 20 minutes per month. And if they hold onto a test that turns out to be prime, it'll be nice to have found it, but in the interim we will have wasted 1/6th of the current CPU power available to SoB. Then again, isn't this what the garbage account is supposed to deal with?
We generally fall back to secondpass if firstpass runs out.
Looking at the levels of n, I think doing either at this point would be fine.
But your right firstpass looks like it will run out in the next two weeks.
Jason
That is what the garbage queue used to be afaik - someone please correct me if I'm wrong.
It would hand out the oldest unreturned WU for which there was no result in the db, which effectively did a doublecheck to make sure it had been tested at least once.
Vato,
I think your correct.
Basically the garbage que could contain those n's as well as any other n's that the higher ups deemed nessary to retest or test.
I guess the major point it that que is populated manually by the project corrdinators.
however it should be checked that only fast computers are give ngarbage tests and make sure eac htest is only assigned once. I believe the old versions tended to assign multiple people the same test.
Concur - suitable logic ought to be something along the lines of:
Populate garbage queue daily with the lowest n outstanding tests (and/or oldest tests outstanding), perhaps with a queue size of ~200.
Remove any unassigned test from the garbage queue if a result for that test is received back from other clients.
Should a result come in during this, it simply means a doublecheck has been done.
I'd be happy to a put a fast core on this to keep things ticking.
(In the meantime - I'm more worried about first-pass not being repopulated before it empties - only 754 tests remain at the time I type this)
And we have new first-pass tests 15M->16M in the queue!
Thanks!
I installed sob on a pc with this "GP500QQQsecondpass" but i don't i got secondpass test. taking too long 110h on 1 coreduo.
can't i force too do secondpasses.
because i tink we are letting the secondpass que build up too much.
Without secondpass there no need for firstpass.
You first need to create a user "GP500" either through the client, or through the website http://www.seventeenorbust.com/
Only then can you use "GP500QQQsecondpass" in the client to force the client to request k n pairs from the second pass queue.
Joe O
I already have been active as that user.
If we look at this
http://www.seventeenorbust.com/secret/
Almost no one is getting secondpass, only 2 per day ??
That looks![]()
I haven't been able to find your stats on this page: http://www.seventeenorbust.com/stats/users/
Joe O
Use caps, It's there.
But anyway why are there so few secondpass jobs, it's important imho.
Nah I think you were thinking of the old stats. Secondpass is at 11.4 m already. On my dual 2.8 xeon it take several days to complete a wu. Bottom line is if you are getting 11m wu, then you are crunching secondpass. If the next wu is 15m then it is possible that you configured your client incorrectly. My second pass client is still getting 11m wu not 15m.![]()
2nd pass here
24737•2^11498407+1 Thu Jul 10 19:04:02 2008 Wed Jul 16 03:55:09 2008 5777730 50 % 1646793
55459•2^11497906+1 Mon Jul 7 21:25:53 2008 Wed Jul 16 04:01:39 2008 9417180 81 % 1741348
55459•2^11497990+1 Tue Jul 8 00:39:07 2008 Wed Jul 16 04:02:51 2008 9243208 80 % 1737202
Taking about 10 days per wu - running 3 wu's on dual core at 2.93Ghz (Pentium D 805 overclocked 10%).