Possible extention to the "GPU to 72 Tool" project?

[chalsall]

Hey all.

First of all, I must thank all of the GPU Workers who have signed up and been using the GPU to 72 Tool. Eighteen workers have, in only a few days, done over 3800 GHz Days of additional TF work at the "LL Wavefront", eliminating 11 candidates (saving approximately 25 needed LL tests).

If anyone who has a GPU and has not signed up yet, please consider doing so.

However, one disappointing thing I've seen is many of the candidates being returned to PrimeNet are being assigned to "Anonymous" for the LL work, rather than known high-producing LL Workers.

So my question is, does anyone think it would be a good idea to rather than return candidates to PrimeNet for random assignment, low candidates instead be made available to certain LL Workers (like those at the top of the LL Producer's list)? (And yes, I know that "Anonymous is number two there, but I also know that "Anonymous" is also used to identify users who never actually finish the work they're assigned.)

How to transfer the candidates to the "trusted" LL Workers is one implementation issue. I can think of two possibilities:

1. The LL Worker is given the "Test=[AID],[exponent],[FactTo],[P-1 Done]" line with the real AID. The Worker then unassigns the work on Primenet and then immediately re-requests it using a "Test=[exponent],[FactTo],[P-1 Done]" which will give the Worker a new AID owned by that PrimeNet account.

1.1. Cons: there is a "race condition" such that some work may be reassigned to another account in between the two steps.

2. PrimeNet is modified to allow an account user to transfer an assignment to another user.

2.1. Cons: this would require work on PrimeNet.

2.2. Perhaps James Heinrich would be willing to help with this, as I believe he now has access to PrimeNet.

Or, can anyone else suggest another methodology?

My above is only a proposal, and is a request for comments. The intention is to compress the "wave front".

Thoughts?

[Mr. P-1]

As best I can tell:

1. A person will get credit if they are logged in to primenet when they submit their results, regardless of who the exponent is actually assigned to.

2. A person whose non-anonymous client checks in with an anonymous assignment key is assigned that exponent without warning. A person whose client check in with an non-anonymous assignment key which is not theirs, is given a warning, but is still assigned the exponent.

Based on these observations, I see no reason for anyone to try to get a new assignment key, unless for some reason they don't trust you.

[chalsall]

Originally Posted by Mr. P-1

As best I can tell:

1. A person will get credit if they are logged in to primenet when they submit their results, regardless of who the exponent is actually assigned to.

2. A person whose non-anonymous client checks in with an anonymous assignment key is assigned that exponent without warning. A person whose client check in with an non-anonymous assignment key which is not theirs, is given a warning, but is still assigned the exponent.

Based on these observations, I see no reason for anyone to try to get a new assignment key, unless for some reason they don't trust you.

Even better! No work needed on PrimeNet, and no race condition.

Now, do you (and others -- particularly George) think this is a good idea?

[garo]

I doubt you'd find enough people to complete all the LL tests.The anonymous assignments is just part of Primenet. One thing we could do is ask george to increase the trusted exponent limit to 48M or more. Another thing is to unreserve the exponents as close to 0000 UTC as possible to allow people who care about this, to snag some exponents.

Still I think you will probably need to keep less than 100 LL in stock.

[petrw1]

Originally Posted by chalsall

However, one disappointing thing I've seen is many of the candidates being returned to PrimeNet are being assigned to "Anonymous" for the LL work, rather than known high-producing LL Workers.
Thoughts?

I have been monitoring the P-1 workers at the LL wavefront to see if they were starting to keep up with LL demand. The available LL exponents had been steadily climbing to close to 4,000 until a few weeks ago when they started to disappear in a big hurry. I queried to see who might have suddenly had the capacity to take so many LL tests ... I was a little surprised to see that the majority of the LL tests in that range were assigned to ANONYMOUS rather than (for example) curtisc as a more likely candidate as the biggest LL contributor.

As I understand the current process, other than Garo's suggestions of raising the preferred assigment limit there is nothing stopping ANONYMOUS or anyone else from grabbing any or all assignments....and as long as some regular progress is reported (no matter how small) they can be retained by ANONYMOUS for up to a full year.

It could be reasonable for your tool to reserve them; up to a limit that known trusted workers can keep up with.

[Dubslow]

I think the feasability alarm needs to be rung, aside from all the other problems. With a top of the line proc, without mfaktc running, I can get somewhere just over 4 LL's done per month, versus hundreds of TF'd exponents per day. Even if we wanted to, we couldn't. As it is, I have around 5 exponents that Mr. P-1 had reserved for this project that also had no P-1; I took 5 to do P-1 and take all the way to completion.

On the other hand, I do like the idea to raise the trusted limit higher, or put any low expiries back into the pool exactly at 0000 UTC. That would make it easier for us to collect as many as we could, even if it's not that many.

[davieddy]

Originally Posted by Dubslow

I think the feasability alarm needs to be rung, aside from all the other problems. With a top of the line proc, without mfaktc running, I can get somewhere just over 4 LL's done per month, versus hundreds of TF'd exponents per day. Even if we wanted to, we couldn't. As it is, I have around 5 exponents that Mr. P-1 had reserved for this project that also had no P-1; I took 5 to do P-1 and take all the way to completion.

On the other hand, I do like the idea to raise the trusted limit higher, or put any low expiries back into the pool exactly at 0000 UTC. That would make it easier for us to collect as many as we could, even if it's not that many.

Good post.

4 LLs per month is good going. But ATM GIMPS completes about 200 a day.
That's 1500 folk like you, or 12000 like me!

As I see it, the main point of TFing as far as possible is to incentivate the great unwashed CPU to embark on and complete LLs.
ATM (I hope temporarily), our efforts are in danger of doing the opposite:
A 55M exponent TFed to 71 doesn't fill me with an irresistible urge to LL it.


More later,
David

[lycorn]

Originally Posted by petrw1

(...) Garo's suggestions of raising the preferred assigment limit (...).

It could be reasonable for your tool to reserve them; up to a limit that known trusted workers can keep up with.

I vote for these 2 suggestions. The first one is pretty obvious and straightforward to implement.
As for the 2nd one, it would be a matter of chalsall´s tool to keep the exponents reserved after the prescribed TF limit (71, 72...) is attained and then release them, upon request, to trusted testers. A way of transferring ownership of the exponent(s) would have to be devised and agreed upon.
Based on what was stated by Mr. P-1 in his last post (point 2.), there seems to be no issues with the Primenet server.

[Dubslow]

Originally Posted by davieddy

Good post.

4 LLs per month is good going. But ATM GIMPS completes about 200 a day.
That's 1500 folk like you, or 12000 like me!

As I see it, the main point of TFing as far as possible is to incentivate the great unwashed CPU to embark on and complete LLs.
ATM (I hope temporarily), our efforts are in danger of doing the opposite:
A 55M exponent TFed to 71 doesn't fill me with an irresistible urge to LL it.


More later,
David

It will in the near future. The most appealing are the lowest ones, of which there are many. There's still >100 tests <40M left.

(Also I only get around 3 per month, what with the mfaktc core, though that is with 53M exponents, so that on 5 the 45M's I have, I might be able to pull off 4 a month on three cores.)

[Dubslow]

Originally Posted by lycorn

I vote for these 2 suggestions. The first one is pretty obvious and straightforward to implement.
As for the 2nd one, it would be a matter of chalsall´s tool to keep the exponents reserved after the prescribed TF limit (71, 72...) is attained and then release them, upon request, to trusted testers. A way of transferring ownership of the exponent(s) would have to be devised and agreed upon.
Based on what was stated by Mr. P-1 in his last post (point 2.), there seems to be no issues with the Primenet server.

Mr. P-1 and I (among others) have tested and used a system where GPU to 72, i.e. chalsall's spider, would collect assignments anonymously; any user with the hex key can then claim the assignment without the risk of it being re-assigned.

The biggest issue is unreserving those exponents we can't take, which will be easily >95% of them.

[chalsall]

Originally Posted by Dubslow

The biggest issue is unreserving those exponents we can't take, which will be easily >95% of them.

Actually, that's not really an issue at all.

The system could simply keep a "store" of a certain number of exponents (say 100) which would have ordinarily been returned to PrimeNet.

As those who participate in the exercise (and it's still a proposal -- I won't implement unless it's agreed it won't have any harm and that it will be used) draw from the pool and take over ownership of the exponents, the system would simply top it up again (read: not release as many exponents back to PrimeNet as it would have ordinarily done).

The Perl and SQL to implement this would actually be less than the English describing it....