Expiration on TF

[9vLYvxWUJ7Dy2Hb]

Hi,


are TF assignments expiring?


Going through https://www.mersenne.ca/status/tf/0/0/1/0 I noticed there are chunks of numbers only factored comparatively poor compared to adjacent areas.


They are assigned since a relatively long time (some Feb./August/July 2020). Given that these are factoring for the 72-75 bit range this seems excessive. Will they eventually expire? and if so, when?



Examples:
https://www.mersenne.org/report_expo...1463987&full=1
https://www.mersenne.org/report_expo...3059631&full=1



https://www.mersenne.org/report_expo...7500103&full=1

[Uncwilly]

What you are seeing is the attempt to do the TF work just before the first time primality checks.
If you look at https://www.mersenne.ca/status/tf/0/0/2/0 the yellow is where we want the TF to be. But there is no point to do too much work too soon. Eventually we do want to go to a higher TF level for the higher ranges. If you look here: https://www.mersenne.ca/status/tf/0/0/4/33200 you will see more has been done in the 332.1M and 332.2M ranges. That is because people are doing testing there for the 100 million decimal digit tests. That is where the next big prize is.

[9vLYvxWUJ7Dy2Hb]

Thanks for the response, not sure I can follow though.

What you are seeing is the attempt to do the TF work just before the first time primality checks.

Are you referring to the TF that Prime95 performs before doing the LL/PRP? These are marked as Work Type PRP etc. though with the Stage TF, right?


The ones I mean list TF as work type so I suppose they were specifically assigned as manual Tasks.



The following gives 224 assignments all assigned at the same date as far as I can see - so these were manual TF assignments, right?

https://www.mersenne.org/report_fact...72&tftobits=72


I understand the chunks of exponents which are particularly well factored because they're particularly interesting due to the 100M digit limit/specific people. But don't really understand chunks which stick out in the other direction (other than, that they were assigned but then forgotten)

[Uncwilly]

Ok, you have brought up multiple issues, not all in alignment with each other. I will try to address everything clearly

Originally Posted by 9vLYvxWUJ7Dy2Hb

are TF assignments expiring?

This can be taken to mean:
Is the TF assignment type going away? No it is still an important work type.
Are TF assignments rotting and failing to complete? Generally not, but some might.
Do TF assignments expire according to some rule? There are no specific assignment rules listed for TF assignments to expire.
https://www.mersenne.org/thresholds/
There should be some rules about TF assignments being allowed in the Cat 0-2 region.

Why no expiry for TF?
Much of the TF happening now is being managed through GPU72. GPU72 is helping us apply the optimum amount of TF in a near "just in time" fashion. This allows more TF and thus fewer primality tests.

Going through https://www.mersenne.ca/status/tf/0/0/1/0 I noticed there are chunks of numbers only factored comparatively poor compared to adjacent areas.

This is in part due to the nature of the way we are doing the TF. We are trying to do it like a wedge ►, with the thickest part happen right before hand of to primality testing. If you look this graph: https://www.mersenne.ca/graphs/facto...M_20201222.png
You will see that the black line is nearly on the green line before the 100M mark. That green line was the goal level for TF work. The black line is the current state of TF work. That green line is the equivalent of the yellow boxes on the link that you referred to. If you click on this link: https://www.mersenne.ca/status/tf/0/0/2/10000 and then this one https://www.mersenne.ca/status/tf/0/0/3/10000 and compare the number in the yellow or above to same ranges here: https://www.mersenne.org/primenet/ you will see that the TF is getting done just ahead of the assignment of First Time checks.

They are assigned since a relatively long time (some Feb./August/July 2020). Given that these are factoring for the 72-75 bit range this seems excessive. Will they eventually expire? and if so, when?

The link that you pointed to above did not give data about assignment dates or any narrowly focused group of exponents. So, I don't exactly understand which you are referring to.

Examples:
https://www.mersenne.org/report_expo...1463987&full=1
https://www.mersenne.org/report_expo...3059631&full=1
https://www.mersenne.org/report_expo...7500103&full=1

These are 'all over the map'. So, looking at these in order of exponent lowest to highest:
103059631 is currently assigned to "GPU Factoring" which is the user ID for GPU72. It is 2 bits lower than the goal bit level of 76. Many other exponent in the 103M range are at that bit level: https://www.mersenne.ca/status/tf/0/0/4/10300
and they are available to people working through GPU72: https://www.gpu72.com/reports/available/
GPU72 will release exponents that it manages as the First Time primality check wave front approaches, so no issue there.
121463987 That is in a range that is 8-18 months ahead of the FT primality checks arriving in that area. If I remember correctly, a user had been working on the 121M range for a long while, so others stayed away. But of late people are working in that range. At the moment SRBase has been feeding BOINC users TF work, going 1 bit level at a time and working above the range that GPU72 is managing. Watch this: https://www.mersenne.org/report_recent_cleared/ (sort it by exponent) over the course of weeks and you will see the progress. If you mouse over the graphic at https://www.mersenne.ca/ and let the animation play, you will see the progress that SRBase is making on the black line in the last year or so. Most all of the further out ranges got brought up to a minimum level and then on the next pass, they went up a bit level.
967500103 is decades away from us getting to there with primality testing. So, worrying about TF testing there is a waste of brain cycles. This again is in a range where some user had been working way ahead of the group for some reason. And since that range was claimed by someone SRBase likely stepped around it. If someone with a GPU wanted to, they could take all of the 967M exponents at 71 and move them up to 72 bits in a week or less. Nothing to worry about.

Originally Posted by 9vLYvxWUJ7Dy2Hb

Are you referring to the TF that Prime95 performs before doing the LL/PRP?

TF is a different work type than LL or PRP and is assigned differently. If you get an assignment for an exponent where you are going to go up by more than one bit level, your assignment is not dropped when you report that you got one bit level done.

These are marked as Work Type PRP etc. though with the Stage TF, right?

No, TF is not a stage of PRP.

The ones I mean list TF as work type so I suppose they were specifically assigned as manual Tasks.

They could be automatic assignments. There is a work type in Prime95 that is "Trial factoring to low limits". This is TF work well ahead of the current area of work. This is mainly a legacy of the time before GPU's were being used to TF.

The following gives 224 assignments all assigned at the same date as far as I can see - so these were manual TF assignments, right?

https://www.mersenne.org/report_fact...72&tftobits=72

That list is showing that they were all assigned today.

I understand the chunks of exponents which are particularly well factored because they're particularly interesting due to the 100M digit limit/specific people. But don't really understand chunks which stick out in the other direction (other than, that they were assigned but then forgotten)

Any range that was claimed and has fallen behind the average for the broader area will eventually get taken to the appropriate level. But, there is no need to worry about things that are months, years, and decades in the future.

[phillipsjk]

So much detail I never fully appreciated.

[9vLYvxWUJ7Dy2Hb]

Wow, thanks for this very informative answer!


It was indeed the "Do TF assignments expire according to some rule?" question I was after. Not really because I deemed it a problem but more because I noticed it looked a little odd and was curious why.

The link that you pointed to above did not give data about assignment dates or any narrowly focused group of exponents. So, I don't exactly understand which you are referring to.

Right - the three examples towards the end of my post were each taken from larger chunks of assignments that had been assigned in the past. - I should have posted the link to the group of exponents instead.

"GPU Factoring" which is the user ID for GPU72.

Ah, interesting I didn't notice that before. I did know the SRbase User (mostly because it's prominently sitting in the first spot for TF)

No, TF is not a stage of PRP.

My Bad, I mixed up TF assignments and the P-1 task that Prime95 does (Or at least it did so in my case before starting the actual PRP)

That list is showing that they were all assigned today.

Right, I noticed that as well - they were assigned since July or October to a user and "updated" in October - until yesterday. Today they are assigned to someone else.

Any range that was claimed and has fallen behind the average for the broader area will eventually get taken to the appropriate level. But, there is no need to worry about things that are months, years, and decades in the future.

Right - the question was mainly curiosity driven. So thanks again for taking the time for this in-depth response!

[Uncwilly]

P-1 is a different way of looking for a factor and should be done before PRP/LL. Traditionally it has been done by the person doing the primality test. But of late, there has been an effort underway to make sure it gets done "well". So a group has been working to do as many as possible on machines with a good amount of memory. More memory available to Prime95 makes for better P-1, thus more factors found and less first time checking.