Strategic Double Clicking

[Mark Rose]

Originally Posted by airsquirrels

All of these triple checks (and the occasional quads) are mostly thankless, but ideally accelerating us towards more strategic double checks by giving us definitive records for more systems. Madpoo would have to give us updated numbers, but there should be under 2000 unassigned triple checks at this point.

I'm tracking 161 unassigned exponents needing TC from this thread that I did not participate in. I've got about 50 DC needing TC, but I don't track how many are assigned.

There are 125 unassigned DC from this thread from 55254491 on up. That's the list I gave to you and MadPoo. I estimate I could complete that work in about six weeks, but I'm working through a queue of about two months already.

[endless mike]

Originally Posted by airsquirrels

http://www.mersenne.org/report_expon...4097311&full=1

Low priority quad check. No other issues with this card in the past. (GTX 690)

Completed. Matched your result.

[airsquirrels]

Originally Posted by endless mike

Completed. Matched your result.

Thanks! What a nice relief.

[Madpoo]

Originally Posted by Mark Rose

I'm tracking 161 unassigned exponents needing TC from this thread that I did not participate in. I've got about 50 DC needing TC, but I don't track how many are assigned.

There are 125 unassigned DC from this thread from 55254491 on up. That's the list I gave to you and MadPoo. I estimate I could complete that work in about six weeks, but I'm working through a queue of about two months already.

The stuff needing triple-checks... well, I'm working on a bunch and so is AirSquirrels.

Besides the ones we've managed to generate from strategic double-checking (where I assume *our* results are the correct ones), there are a bunch more.

If people are really anxious to do the necessary triple-checks, I can post a big list on here and let you pick your poison, as long as it wasn't one where you did one of the previous checks of course.

There were 2000+ unassigned of those, I think. The churn in the 40M-45M range is where a bunch of new stuff seems to show up weekly, and I expect that's normal... of all the double-checks being turned in daily, I guess I'd expect 4-5% of those to mismatch and need a triple check (given the average historical error rate).

A large % of the mismatches come from the first result being suspect, so it was handed out pretty quick for another "first time" check. Maybe 80% of the mismatches are like that, give or take.

The ones where neither result was suspect are interesting, since it means some machine was spewing bad stuff and was blissfully unaware of it, but I haven't found resolving those to be any more or less useful in finding unknown bad systems.

Anyway, my own little goal is to hopefully clear out the backlog of these mismatches and maybe, or maybe not, keep pace with the new ones that come in. They may come in faster than I can test them... who knows, but it's kind of fun to essentially pick the winner and loser in the contest, as tiebreaker. LOL

[bgbeuning]

This has probably been explained in the 100 pages of previous posts on this thread,
but I am too lazy to read them all. Why doesn't primenet keep assigning an exponent
until it gets two residue matches from different users? So the triple checks could be
handled by the entire pool of prime95 users. Not really complaining but so much is
automated already why not this little extra part also?

[endless mike]

Originally Posted by bgbeuning

This has probably been explained in the 100 pages of previous posts on this thread,
but I am too lazy to read them all. Why doesn't primenet keep assigning an exponent
until it gets two residue matches from different users? So the triple checks could be
handled by the entire pool of prime95 users. Not really complaining but so much is
automated already why not this little extra part also?

As far as I know, it does. Some of us are just impatient.
I once was staring at approximately 60 mismatched exponents on my results page. It's down into the teens now. A lot of them were from doing strategic double checking from this thread. It would have taken quite some time for the double checking to catch up to them and give confirmation.

[airsquirrels]

Originally Posted by endless mike

As far as I know, it does. Some of us are just impatient.
I once was staring at approximately 60 mismatched exponents on my results page. It's down into the teens now. A lot of them were from doing strategic double checking from this thread. It would have taken quite some time for the double checking to catch up to them and give confirmation.

I think that is the main purpose for a lot of us doing triple checks. Generally we're running faster, dedicated checks than the churn that would normallly pick them up. Perhaps making all triple checks the equivalent of DC Cat 0 would have the same effect of ensure users with hardware trouble know as soon as possible.

It is also nice to work on an obtainable subgoal vs. just keep going on the decades of checks remaining in the overall queue.

[GP2]

Originally Posted by bgbeuning

This has probably been explained in the 100 pages of previous posts on this thread,
but I am too lazy to read them all. Why doesn't primenet keep assigning an exponent
until it gets two residue matches from different users? So the triple checks could be
handled by the entire pool of prime95 users. Not really complaining but so much is
automated already why not this little extra part also?

I think that's exactly what it does.

As far as I know, PrimeNet doesn't treat triple checks any differently than double checks. They get assigned "in due course". The problem is that the overwhelming majority of PrimeNet users are anonymous churners, who abandon exponents before completing them (or in many cases, even starting them). Eventually the exponent expires, and often it gets assigned to another anonymous churner, and the cycle repeats. Or sometimes the exponent does run to completion, but on a very slow machine.

Collectively the churners make a very important contribution, since there are so many of them, and some percentage of exponents do run to completion. However, any individual exponent might get churned half a dozen times or more over the course of many months or even a few years.

Madpoo et al. seek to short-circuit this process by manual assignment of triple checks to themselves, to get them done with higher priority. Maybe there could be some automation of this process, maybe PrimeNet could identify triple checks and classify them as Category 1 and hand them out to people who agree to do them, or something like that. But probably when the existing backlog is cleared out it will be easy to keep up with the pace of creation of new triple checks and keep on handling them manually.

[Madpoo]

Originally Posted by GP2

I think that's exactly what it does.

As far as I know, PrimeNet doesn't treat triple checks any differently than double checks. They get assigned "in due course". The problem is that the overwhelming majority of PrimeNet users are anonymous churners, who abandon exponents before completing them (or in many cases, even starting them). Eventually the exponent expires, and often it gets assigned to another anonymous churner, and the cycle repeats. Or sometimes the exponent does run to completion, but on a very slow machine.

Collectively the churners make a very important contribution, since there are so many of them, and some percentage of exponents do run to completion. However, any individual exponent might get churned half a dozen times or more over the course of many months or even a few years.

Madpoo et al. seek to short-circuit this process by manual assignment of triple checks to themselves, to get them done with higher priority. Maybe there could be some automation of this process, maybe PrimeNet could identify triple checks and classify them as Category 1 and hand them out to people who agree to do them, or something like that. But probably when the existing backlog is cleared out it will be easy to keep up with the pace of creation of new triple checks and keep on handling them manually.

^^^ Yes, that.

I kind of like doing the triple-checks because I know when some people do a double-check and it mismatches, they start to wonder "is my machine bad?" and some of them are tempted to do their own additional test.

I think it's weird to have people verify their own work (there was that whole project to "fix" those) so this is kind of my way of helping out.

I recently recall going through a bit batch of endless_mike's mismatches (I wasn't picking those out on purpose, it just happened to fall in the range I was doing at the time) and I'm sure he appreciated knowing for sure his were correct.

It would be kind of cool to have a "triple check" option for getting assignments... I guess if the pool of suitable work were large enough that would be worth a long term effort, but ideally we can get the backlog cleared out soon (end of year perhaps, if it's just AirSquirrels and myself? Sooner if others want to help?)

By happy coincidence, most of the new mismatches are smack-dab in the middle of the cat4 DC range since that's where most new results are showing up. That makes it more likely that they'll be reassigned automatically anyway...

Eventually, once this backlog is done, I foresee most of the manual effort being the cleanout of the occasional mismatch in the higher ranges (when some slow, expired assignment turns in *after* the newly assigned worker turns theirs in), or the stuff in cat2/cat3 which doesn't get handed out as swiftly.

[Madpoo]

By the way, if anyone feels like doing a quad-check on something, these exponents are currently unassigned. I did one of the previous checks on each of them so I'm ineligible.

41931817
40788089
43125469
44529083
59999201
60025513
61433279

[rudi_m]

Originally Posted by Madpoo

By the way, if anyone feels like doing a quad-check on something ...

I took them all.