mersenneforum.org > Data Ahmerali
 Register FAQ Search Today's Posts Mark Forums Read

 2005-11-22, 18:20 #1 lycorn     "GIMFS" Sep 2002 Oeiras, Portugal 17·89 Posts Ahmerali Just noticed we´ve had a turbocharged user joining our ranks. They came out of the blue (at least I had never noticed them before) and are in 16th place in the LL test ranking. Anybody knows how many machines they are using?
 2005-11-27, 05:34 #2 ahmerali     Jul 2005 1316 Posts Sure, I can tell you, since I'm him. About 475, several of them hyper-threaded. Can't say too much more...
 2007-02-03, 20:57 #3 Joe O     Aug 2002 10000011012 Posts Congratulations on reaching 3rd place!
2015-05-11, 20:53   #4
Serpentine Vermin Jar

Jul 2014

31×107 Posts

Quote:
 Originally Posted by ahmerali Sure, I can tell you, since I'm him. About 475, several of them hyper-threaded. Can't say too much more...
I hate to revive a decade old post, but since I was looking at stats on computers (especially from heavy contributors) with a larger than usual "bad result" percentage, I dug into who this "ahmerali" user is.

Besides getting stats on the % of bad results, I'm also doing triple-checks where the same user did the first and double-check and "ahmerali" came up a lot in those as well. No more so than Curtis or other large contributors so it could just be due to the large quantities.

Anyway, "ahmerali" currently has:
Code:
result_state	Count
Unverified	3137
Verified	3685
Suspect	69
Factored	34
The "suspect" results are when 2 checks returned 2 different residues. The one with a non-zero error code is marked suspect in those cases... a verifying triple-check will either put that into the "verified" or "bad" column.

So far, that puts the ratio of bad/verified at 5.2% which is kind of high. There are those other 69 that have been double-checked without a match, and ahmerali's results had a non-zero error code during the run.

Many of the 3137 unverified results have not been double-checked yet... some have though and there wasn't a match so ahmerali's results weren't marked suspect, but I wonder. It's unusual to have so many mismatches in the unverified column that need a triple-check now.

To be precise, 64 of those 3137 in the unverified column need a triple-check (or even a quadruple check in the case of http://www.mersenne.org/M35205953

I don't know if ahmerali is still around... doesn't seem to have contributed results lately, but I did wonder if there were any known problems with any of the machines? Something that could help me narrow down certain ones that had more bad results than others?

These came from v4 systems and unfortunately on the v5 system all of the different computers got lumped into a single "v4 computers" id. I may try to pull apart the old v4 info and see if it was just a particular machine and try and get those triple-checked (or do a verifying check) sooner than the double-check wavefront will hit it.

2015-05-11, 21:22   #5
Serpentine Vermin Jar

Jul 2014

31×107 Posts

Quote:
 Originally Posted by Madpoo These came from v4 systems and unfortunately on the v5 system all of the different computers got lumped into a single "v4 computers" id. I may try to pull apart the old v4 info and see if it was just a particular machine and try and get those triple-checked (or do a verifying check) sooner than the double-check wavefront will hit it.
Ah, I see how the v4 results are actually included in the data... never mind on that, I can query using some additional info now. So far it does look like a majority of the bad results did come from a few bad systems. One of them accounted for 189 bad results alone. The other results from that machine have all already been verified or triple-checked, so there's nothing left hanging there. I'll work through some other stuff.

The total # of exponents involved is probably higher too if I include some of the old v4 results that weren't being included previously. Chalk that up to weird things during v4/v5 migration I guess.

EDIT: Yup, when including some of that extra v4 data, it's 547 bad out of 9708 verified. Makes the bad % a little higher in fact at 5.6%.

Well, project for another day I guess. I still have lots of triple-checking going on.

Last fiddled with by Madpoo on 2015-05-11 at 21:36

 2015-05-12, 03:33 #6 LaurV Romulan Interpreter     "name field" Jun 2011 Thailand 24×613 Posts He said he hyperthreads and overclocks and uses lots of machines without discrimination, so in this case, the 5% error may be quite normal, in spite of the fact that it looks like cheater, for me. But again, I am the paranoid one here..
2015-05-12, 17:54   #7
Serpentine Vermin Jar

Jul 2014

63658 Posts

Quote:
 Originally Posted by LaurV He said he hyperthreads and overclocks and uses lots of machines without discrimination, so in this case, the 5% error may be quite normal, in spite of the fact that it looks like cheater, for me. But again, I am the paranoid one here..
Once I'm further along with some of these other triple-checks I may devote some resources to doing double-checks on his work (and other systems that seem to have a higher than usual error rate).

I actually picked up a batch of maybe 15 or so exponents in the 35-37M range that were awaiting a needed triple-check because of mismatched residues. As expected, the user "MrRoboto" has a very high failure rate... so far we're at 29 verified and 22 bad with another 21 of his results in an unverified state, and most of those have already had a mismatched double-check... I'm guessing 20-21 of those will also turn out to be bad.

All times are UTC. The time now is 21:38.

Tue Dec 7 21:38:26 UTC 2021 up 137 days, 16:07, 1 user, load averages: 1.93, 1.40, 1.35