[QUOTE=Gordon;414715]Never a dead end until you actually do it and get the answer...it's really just about neatness, so when you look at the [URL="http://www.mersenne.ca/status/tf/0/0/4/0"]progress chart[/URL] eventually everything will be at least 64 bits.[/QUOTE]

Sure... Burning a few dinosaurs to make things look "neat" to the very few who look at the report makes a great deal of sense....

[QUOTE=chalsall;414719]Sure... Burning a few dinosaurs to make things look "neat" to the very few who look at the report makes a great deal of sense....[/QUOTE]

Yep, my money, my equipment :smile:

[QUOTE=Gordon;414721]Yep, my money, my equipment :smile:[/QUOTE]

Yup. Rock your boat! :smile:

[QUOTE=Gordon;414718]
now scroll down to the entry for M16649 and you will see the following

T25 -Done
T30 - Done
T35 - Done
T40 - Done
T45 - 1864
T50 -

So that means that this exponent must have had 9064 curves run right?
Why can't the report show the actual number of curves run at each level as I have summarised above, you can then use this to plan where to put the effort in.[/QUOTE]

1. No, 9064 curves do not need to be run for this status to be correct. Those 100 curves at t50 level are equivalent to an entire t35 on their own, as one example. Have you noticed how the moment, say, a t35 is complete that there's hundreds of t40-level curves already listed? That's because the 1580 curves at 1M are *equivalent* to some number of curves at t40. The PM1 work that is done also counts as ECM work, which may explain the "missing" curves when taking into account equivalences. One curve at 44e6/6e9 is worth almost 4 default t45 curves.

2. The report does not show the actual number of curves run at each level because it would confuse the people who do not understand ECM well, while causing unneeded headaches for the people who do understand ECM. Even if the report is slightly optimistic, running curves at higher bounds rarely wastes much effort while running curves at too low a bound wastes quite a bit of time.

[QUOTE=VBCurtis;414739]1. No, 9064 curves do not need to be run for this status to be correct. Those 100 curves at t50 level are equivalent to an entire t35 on their own, as one example. Have you noticed how the moment, say, a t35 is complete that there's hundreds of t40-level curves already listed? That's because the 1580 curves at 1M are *equivalent* to some number of curves at t40. The PM1 work that is done also counts as ECM work, which may explain the "missing" curves when taking into account equivalences. One curve at 44e6/6e9 is worth almost 4 default t45 curves.

2. The report does not show the actual number of curves run at each level because it would confuse the people who do not understand ECM well, while causing unneeded headaches for the people who do understand ECM. Even if the report is slightly optimistic, running curves at higher bounds rarely wastes much effort while running curves at too low a bound wastes quite a bit of time.[/QUOTE]

So are you saying that the 915 curves actually run equate to completing fully T25, T30, T35, T40 and nearly 2000 curves at T50?

Seriously?

Just to hammer the point home about how what is shown makes no sense at all, let's revisit M1277, I ran another 50,000 curves with B1=50K, checked the result in and the T65 count went up by 3.

What are the chances (actual %) of 50K T25 curves actually finding a 65 digit factor?

If I was so inclined within a reasonably short period of time (few months) I could make M1277 indicate that T65 was "complete" and anyone looking at that report to pick exponents to factor would likely skip over it.

I see no downside to my suggestion, if you fully understand it you can do the comparison in your head, for those that don't - or don't want to and just do the work - they would quickly be able to see what is needed.

[QUOTE=Gordon;414804]So are you saying that the 915 curves actually run equate to completing fully T25, T30, T35, T40 and nearly 2000 curves at T50?[/quote]
The 2000 curves are listed against t45 level, not 50. Nonetheless, given the results data, it should be about 151+4*100 = 551 (+ few extra) curves at t45 level. I would think that this is due to missing results data rather than incorrect summation. Primenet sums up the ECM results submitted to show the various counts. I'd trust that statistics more than the results data, especially since there doesn't appear to be any ECM activity on these exponenets in the past 5 years (very suspicious).

[QUOTE=Gordon;414804]Just to hammer the point home about how what is shown makes no sense at all, let's revisit M1277, I ran another 50,000 curves with B1=50K, checked the result in and the T65 count went up by 3.

What are the chances (actual %) of 50K T25 curves actually finding a 65 digit factor?

If I was so inclined within a reasonably short period of time (few months) I could make M1277 indicate that T65 was "complete" and anyone looking at that report to pick exponents to factor would likely skip over it.[/QUOTE]
Running 3 curves at t65 is roughly the same effort as running 50k curves at t25. To finish up t65 that way would require you to run 6 billion curves at t25 level. Which has a pretty decent chance of actually finding a t65 if one exists. But it will not be nearly as efficient as directly running t65 level curves. I say go for it.

[QUOTE=Gordon;414804]So are you saying that the 915 curves actually run equate to completing fully T25, T30, T35, T40 and nearly 2000 curves at T50?

Seriously?

I see no downside to my suggestion, if you fully understand it you can do the comparison in your head, for those that don't - or don't want to and just do the work - they would quickly be able to see what is needed.[/QUOTE]

1. No, I didn't think we were talking about t50 level. Mersenne.org lists 1864 curves at t45 level at present. If we give about 30% credit for curves at 3M and 4x credit for 44M, I count roughly 750 curves equivalent. The PM1 work may be in the vicinity of 50 curves, or more; I am not sure of the conversion for mersenne special form (I think PM1 is more likely than a regular ECM curve to find a factor, due to form 2kp+1 of factors). That means roughly half the indicated work is accounted for in the detailed listing. Running curves at 3M, as it appears you did, is a waste of time compared to running them at 11M or larger.

I've no idea what work is missing from the detailed report, and I do wonder how much work the admins think is missing from the detailed reports.

2. How would a listing of actual curves completed allow someone like you to see what is needed? Let's continue with M16649 as the example. With the curves listed on the detailed report now, ignoring the chart listing 1864 curves at t45 level right now, what is needed? Is a t40 complete? If not, how many more curves are needed?

I agree that 50k curves at B1 = 50k is insignificant for a t65 (not nearly 3 curves at 800M). 230 billion (per GMP-ECM, I assume you're using that for M1277) such curves would be a t55, and still useless for t65. This does make me question the site's algorithm when upconverting small curves.

[QUOTE=VBCurtis;414826]I agree that 50k curves at B1 = 50k is insignificant for a t65 (not nearly 3 curves at 800M). 230 billion (per GMP-ECM, I assume you're using that for M1277) such curves would be a t55, and still useless for t65. This does make me question the site's algorithm when upconverting small curves.[/QUOTE]
If I recall correctly, Mersenne.org uses a very primitive way to calculate the total ECM effort, by converting the ECM curves to GHzdays.
So:
50,000 curves with B1=50,000 B2=5M => 0.85 GHzdays
3 curves with B1=800M B2=80G => 0.816 GHzdays

[QUOTE=VictordeHolland;414829]If I recall correctly, Mersenne.org uses a very primitive way to calculate the total ECM effort, by converting the ECM curves to GHzdays.[/QUOTE]

Yes, it is primitive but has nothing to do with GHz-days.

Using a formula from Alex Kruppa your B1=x B2=y curves=z values are converted into an equivalent number of curves where B2=100x (call this z1). Then z1 * x is added to the running total of ECM effort. I'm no expert in the field, but I've been told this is good enough for a rough approximation of effort.

The most likely cause for the history report not matching the actual total is someone reporting to me by email a substantial number of GMP-ECM results. I have two ways to add this to the database. 1) Convert it to a prime95 compatible format and use the manual web forms , or 2) use a SQL stored procedure that adds to the ECM effort. Method 1 creates a history entry, method 2 does not.

[QUOTE=axn;414807]The 2000 curves are listed against t45 level, not 50. Nonetheless, given the results data, it should be about 151+4*100 = 551 (+ few extra) curves at t45 level. I would think that this is due to missing results data rather than incorrect summation. Primenet sums up the ECM results submitted to show the various counts. I'd trust that statistics more than the results data, especially since there doesn't appear to be any ECM activity on these exponenets in the past 5 years (very suspicious).


Running 3 curves at t65 is roughly the same effort as running 50k curves at t25. To finish up t65 that way would require you to run 6 billion curves at t25 level. Which has a pretty decent chance of actually finding a t65 if one exists. But it will not be nearly as efficient as directly running t65 level curves. I say go for it.[/QUOTE]

Sorry my mistake I did of course mean T45 :blush:

Still, given the results data visible in the database it just doesn't add up.

If this report isn't pulling the data from the database and summing it, where is it getting it's information from ?

Added ; typed the above as George was posting his reply...

[QUOTE=Prime95;414836]Yes, it is primitive but has nothing to do with GHz-days.

Using a formula from Alex Kruppa your B1=x B2=y curves=z values are converted into an equivalent number of curves where B2=100x (call this z1). Then z1 * x is added to the running total of ECM effort. I'm no expert in the field, but I've been told this is good enough for a rough approximation of effort.
[/QUOTE]

that doesn't sound quite right, lets look at my recent 16553 results

B1=3M = x
B2 = 500M = y (167xB1 not 100xB1)
Curves=533 = z

The figures for B1, B2 & curves were arrived at by testing as follows

B2=50M - time a curve under ecm and note how many curves are required
B2=100M - time a curve under ecm and note how many curves are required
and so on

Knowing how long P95 takes to do stage 1 with B1=3M you can calculate the total time required to complete say T35

Find the combination of B2 & curves that gives the lowest total time.

For 16553 it goes like this

B1=1M, lowest total time is when B2=70M needing 1780 curves, total time 20.47 hours. curves required provided by gmp-ecm

B1=3M, lowest total time is when B2=500M needing 533 curves, total time 19.24 hours

B1=11M, lowest total time is when B2=1B needing 231 curves, total time 24.87 hours

Per the formula

1. Throw away Z - the actual number of curves - we don't use it !!
2. Force B2=100*B1 = 300M
3. Calculate Z1*x = 3M*300M = 900*M*M = 9000B
4. ???