Time to change LL vs. TF weight?

[GP2]

Originally posted by nfortino
The trailing edge of TF is 20.1M, not that of LL.

Oops, OK.

But by that definition, the trailing edge of LL testing is at 10.4M.

The main issue is, there's no shortage of exponents available for LL testing at the leading edge... only about 1.5% of exponents there are unavailable due to still being trial-factored.

Perhaps the new v5 server will start handing out a few double checks to accounts that select 'whatever work makes most sense'... close up the gap between DC and LL a bit, and give TF a little more breathing room. Two birds with one stone. But at the moment, it doesn't look like there's cause for alarm.

[nfortino]

Originally posted by GP2

Perhaps the manual range 20104000 to 20150000 (listed as '***** not assigned to anyone!!! ***') should be in fact be released to PrimeNet.

Would it be possible to publicly release the manual reserved ranges some place like opteron? (without the names would be fine if that's the problem) Also, having the manual ranges intertwined with the server ranges causes problems, or at least confusion (I believe it is the cause of the subject of this tread). It may make more sense to give manual ranges far from those on PrimeNet.

[PrimeCruncher]

Originally posted by GP2
At the current leading edge of LL testing, we're waiting for only about 1.5% of the exponents to complete TF -- about 30 out of 2000 -- which leaves the other 98.5% available for LL testing. That's not a shortage.

Ah. But from the leading edge of LL to trailing edge of TF there are over 600 TFs still to be completed. In my opinion, that's too many.

There won't be a shortage of LL tests unless the leading edge of LL testing catches up to the leading edge of TF. But at current rates of convergence (http://opteron.mersenneforum.org/png/leading_edge.png) that won't happen for many years.

That's true but if we ignore the problem it will get worse and then in a few years we will have to deal with it. I say it's better to deal with it now rather than wait and cause everyone to have headaches over the problem.

[dsouza123]

Does the server only give out a number for LL testing once it has been TFed ?

Maybe something like the method of giving LL or DC or TF depending on CPU speed could be further refined for TF.

For AMD Athlons, Durons, and perhaps Intel P3s which are fast at TFing for 64 bits and below, a mini TF could be given which allows a quick factoring to 64 bits (inclusive) to be done and the leading TF could move ahead more quickly.

[nfortino]

Originally posted by dsouza123
Does the server only give out a number for LL testing once it has been TFed?

Except for numbers over 10,000,000 digits, yes.

Mildly off topic, does anyone know what the deal with 17883403 is? In status.txt, it is shown as being trial factored, but not in the summary.

[GP2]

The latest graph of leading edge shows a post-M40 uptick in the first-time LL testing:
http://opteron.mersenneforum.org/png/leading_edge.png

These are just assigned exponents... no guarantee that they will run to completion.

More detail is shown in the attached graph below. Note: the range 21.9M is entirely reserved for manual testing, so it was skipped over, which increases the jump.

[lycorn]

I think we shouldn´t be alarmed by the recent boost observed in first-time LLs. In fact, this is the M40 effect and (unfortunately ...) most of these new assignments will not be completed. Over the next month, we will see many of these exponents expire, holding back the progress of the leading edge, and things should get back to cruise speed shortly after. It means that TF will stay clear of LL. No worries !
I would anyway reiterate the suggestion already made that DCs should always be the first assignments for any new machine, as a way of closing the gap between LL and DCs and of providing new users with faster feedback. Now it can be added that it will also help holding back the leading edge of LL.

[eu-Robert]

Apparently the majority of new users doesn't even finish their first assignment, usually a LL-test. If they don't see a first result after, say, a week, a lot of them will simply give up, never to return.

Do we want more new users to stick with the project? I think we do. We could use all the help we can get. But is it reasonable to expect them to wait for weeks before seeing their first result? I don't think so.

So, I'd suggest, let the first three assignments new users get be TF assignments. Then allow them to do a doublecheck, and only after that, they can choose the type of assignment they want.

I'd be interested in hearing your views on this...

Happy prime hunting,
Robert van der Peijl.

[PrimeCruncher]

Originally posted by eu-Robert
Apparently the majority of new users doesn't even finish their first assignment, usually a LL-test. If they don't see a first result after, say, a week, a lot of them will simply give up, never to return.

Do we want more new users to stick with the project? I think we do. We could use all the help we can get. But is it reasonable to expect them to wait for weeks before seeing their first result? I don't think so.

So, I'd suggest, let the first three assignments new users get be TF assignments. Then allow them to do a doublecheck, and only after that, they can choose the type of assignment they want.

I'd be interested in hearing your views on this...

Happy prime hunting,
Robert van der Peijl.

I tend to agree. LL tests typically take weeks if not months. New users need to see a rise in stats ASAP after coming on board. I think a few TFs followed by a double-check should be recommended but not required, so that if a new user wants to begin hunting for primes immediately, they can do so. We'll just warn them with a pop-up how long they can expect a Lucas-Lehmer test to take.

[outlnder]

The default work unit could be a TF instead of what makes the most sense.

[PrimeCruncher]

A TF does make the most sense for a new user, therefore it should be default anyway.