LL Testing - Smallest Exp vs Standard

[Fred]

When doing standard first time LL testing, its seems that exponents that are assigned currently are in the vicinity of 76M. If requesting a manual "smallest available" first time test, or if you have your account set to "get the smallest exponents", the numbers are in the 67M vicinity.

I'm curious about the reason for the distinction between the two. I get that some folks may only want to work on "World Record" exponents, but there is a separate category for that.

I guess my question is why a request for first time LL exponents defaults to exponents in the 76M range when lower ones have still not been tested. Just curious.

[Mark Rose]

The Assignment Rules page at http://www.mersenne.org/thresholds/ explains most of that. :)

[Madpoo]

Originally Posted by Fred

When doing standard first time LL testing, its seems that exponents that are assigned currently are in the vicinity of 76M. If requesting a manual "smallest available" first time test, or if you have your account set to "get the smallest exponents", the numbers are in the 67M vicinity.

I'm curious about the reason for the distinction between the two. I get that some folks may only want to work on "World Record" exponents, but there is a separate category for that.

I guess my question is why a request for first time LL exponents defaults to exponents in the 76M range when lower ones have still not been tested. Just curious.

There are people (I guess I'd have to include myself) who like to see the milestones moving along without getting bogged down by some small exponent that got assigned to a slowpoke.

So the "smallest available" exponents is setup to try and keep a single slow assignment from gumming up the works.

In theory it sounds good, but in practice it's still all too possible for a slowpoke user to let an assignment wither, so it goes the full expiration threshold before being reassigned. It's the reason we sometimes get several milestones all cleared at once, where a couple different first time milestones are completed in one day, sometimes even 3 at once.

We have some folks on the milestone thread chit-chatting about possible changes, like figuring out which systems are actually super-duper awesome and just give them the smallest stuff automatically, no more opt-in.

I'm actually brainstorming possible SQL solutions to figure out, on the fly, if a system is awesome or not based on the past XX months of data from it... it has to perform well since getting assignments right now is currently pretty top-heavy anyway (goes through a whole song and dance for each exponent being returned, rather than one request that can return multiple assignments at once).

Who knows, by the time I'm done I may find that I've rewritten the whole assignment process in a different way that requires code changes to the website. But it's my day off and it's kind of fun, so what the heck.

[Fred]

Originally Posted by Madpoo

There are people (I guess I'd have to include myself) who like to see the milestones moving along without getting bogged down by some small exponent that got assigned to a slowpoke.

I see. So is the following correct?

By default, first time LL work is based on an incremental counter, and that counter only goes up (progress moving forward), and the counter is currently somewhere in the 76M area?

But, a proven reliable machine which gets 67M exponent is getting it because the account is set to smallest exponents and that is the smallest available exponent that doesn't yet have a confirmed result (ie- it was assigned, probably a year or two ago, but a valid result was never returned).

Is that about right?

[ATH]

Here you can see the lowest exponent that has no LL test is 63,349,229:

http://www.mersenne.org/report_milestones/


Here you can see the different categories:

http://www.mersenne.org/thresholds/

So Cat 1 right now is from 63,349,229 to 67,655,202 and they are given to those that actively changed their settings to lowest available exponents and if their computers live up to the requirements listed next to Cat 1.

Cat 2: 67,655,202 to 68,666,700 are also assigned to people that changed their settings manually but where their computer only lives up to the Cat 2 requirements and not Cat 1.

Cat 3: 68,666,700 to 75,784,772 are assigned to computers that live up to the Cat 3 requirements but not Cat 1/2 and have actively changed settings.

Cat 4: 75,784,772+: People who have not changed settings and computers that does not live up to the requirements

These category limits are calculated every day at 11:20 pm UTC based on the first 4,000/10,000/100,000 exponents that have not been LL tested.

[chalsall]

Originally Posted by ATH

Cat 3: 68,666,700 to 75,784,772 are assigned to computers that live up to the Cat 3 requirements but not Cat 1/2 and have actively changed settings.

Just to be exact, Cat 3 will be given to anyone who use the Prime95/mprime client to fetch candidates and said client has returned at least two results per worker thread in the last 210 days.

You only need to "promise" for Cat 3 if you're getting Manual Assignments.

[Fred]

Awesome, thanks for all that info gents. I had been through all the web pages, including the assignment rules, but it was still a bit fuzzy. I have a much better understanding of it now. Thanks again.

[Madpoo]

Originally Posted by Fred

Awesome, thanks for all that info gents. I had been through all the web pages, including the assignment rules, but it was still a bit fuzzy. I have a much better understanding of it now. Thanks again.

Yeah, it's a little strange to have the option to push a button and get lowest-possible stuff be over on that "assignment rules" page instead of somewhere like the account options where you can change display name, etc.

It could be worth trying to either move that or let you set the option there as well, which is sort of where I'd expect it to be if I were visiting the site for the first time.

Maybe that's why not as many people have opted in for that, because it's harder to find. All the more reason I wouldn't mind just automatically giving lowest-exponents to machines that do have a solid track record, whether they ask for them or not.

After all, I think getting the smallest exponents is something that people like to get because they don't take as long to test, so it's kind of a nice reward for systems that have proven to be reliable, consistent, etc. And the type of thing anyone can take advantage of once they've turned in a few results over a few months, no extra steps needed.

[endless mike]

Originally Posted by Madpoo

It could be worth trying to either move that or let you set the option there as well, which is sort of where I'd expect it to be if I were visiting the site for the first time.

I've been wondering if it could be set on a machine specific basis instead of account wide.

[S485122]

The "get the lowest stuff" setting can be set by computer : just ensure that the days of work to queue up are within the limits :

Cat 1 : "days of work to queue up" <= 10
Cat 2 : "days of work to queue up" <= 30
Cat 3 : "days of work to queue up" > 30

Jacob

[Madpoo]

Originally Posted by endless mike

I've been wondering if it could be set on a machine specific basis instead of account wide.

Technically it is done by computer as well. That is, you enable that option for an account, but only eligible cpus under that account will benefit from it.