Index 1 Sequence Work for the "n^i" Aliquot Project

[EdH]

Originally Posted by EdH

Yeah, as I rediscovered, my ecmpi cluster is much more efficient than my GPUs. 56^97 just broke during t45. I've got all the 150 and 151 composites queued for the night. They should do up to t47.5 according to the wraithx calculator, which equates to about an hour if no factors are found. Let's see what may turn up.

This should have been in this thread. (I may move it someday.)

In any case, I only got one extra of the list to factor and it was done during the t45 step, so the extra 40 minutes to get to t47.8 did not reap any benefit. I've scaled the ECM back to t45 and will work there for the rest (at least for now). I currently have all the c152 and c153 queued to see how they turn out today.

[gd_barnes]

There appear to be 8 splits here of cofactors <= 156 digits as follows:

Opposite parities, no double square bases:
56^97
75^92
77^90
95^78

Same parities or double square bases:
69^91
200^77
696^60
966^54

An impressive haul! :-)

On the opposite parities, all of the latest indexes are ECM'd to t35. Status of the same parities is in the "somewhat easier" thread.

[EdH]

About half of the above shed a 4x factor and remained on index 1, so I had to run the remaining cofactor via GNFS. They were simple enough: 103, 11x, etc. It gave my non-ecmpi machines something to do.

[VBCurtis]

I'll run 94^100 C156.
I marked it in post 1.

[EdH]

Great - Thanks!

[EdH]

I made it through all but the last two with t45. I've started back down from those two at t50. I'm not sure how long my attention span will allow me to continue, but we'll see.

[gd_barnes]

Originally Posted by EdH

I made it through all but the last two with t45. I've started back down from those two at t50. I'm not sure how long my attention span will allow me to continue, but we'll see.

Can you be more specific?:

1. Did you ECM the entire list to t45 with the exception of the two at the end, i.e. 95^94 and 86^99?

2. Where have you started t50 from? Did you do 95^94 and 86^99 to t50 and then start again from the top doing everything to t50 until you're ready to call it quits?

I feel like at C<=151 would not need any more than t45 and then gradually increasing after that to t50. Maybe your process shows differently.

[gd_barnes]

Since my last posting of splits here, there have been 8 more:

Opposite parities, no double square bases:
54^99
78^89

Same parities or double square bases:
68^100
69^97
86^96
120^80
200^73
392^64

The t45 effort has been a fruitful one!

Like before, on the opposite parities, all of the latest indexes are ECM'd to t35 and the status of the same parities is in the other thread.

[EdH]

Sorry for my lack of clarity. (It made sense to me.)

What I did was to start working backwards up the list. That took the last two into the t50 and gave me the idea that I could get the harder ones out of the way. My list became:

86^99: 192/189
95^94: 186/186
162^90: 200/184
78^98: 186/184
162^85: 189/183
102^92: 186/183
120^95: 198/182
92^97: 191/182
229^79: 185/181
119^94: 195/181
99^99: 198/181
96^94: 187/181

I've completed these with the final (96^94) turning up a p52. I was about to decide against continuing after 11 empty runs of between 2.5 and 3 hours, depending on the machines in the cluster for that run. The success was after only 15:47, rather than 179:53 for the previous.

Now I'm thinking of running some more in the same fashion, from the larger toward the smaller. I agree that c15x would probably be overkill at t50, which was why I decided to go from larger toward smaller and cut it off when I got down there, if my patience allowed even that.

I'm glad you consider the t45 effort "fruitful." I felt they were sparse, at the time of each. But, the number of index 1 sequences left is dwindling slightly. That's good to see.

[gd_barnes]

Originally Posted by EdH

Sorry for my lack of clarity. (It made sense to me.)

What I did was to start working backwards up the list. That took the last two into the t50 and gave me the idea that I could get the harder ones out of the way. My list became:
<snip>
I'm glad you consider the t45 effort "fruitful." I felt they were sparse, at the time of each. But, the number of index 1 sequences left is dwindling slightly. That's good to see.

Well...that is bizarre. :-) There's no way I would have read that as you starting to work backwards from the largest cofactors. What threw me is that you said you "started back down" making me think that you started back from the smallest cofactors again. It feels like you're working up the list to me since it's sorted ascending. Anyway, thanks for the clarification. Funny how things can be so easily misunderstood, especially on the internet.

By my observation, I found that 8 sequences were split two different times per my two posts. So that's 16 out of ~160. ~10% is not bad IMHO since I ran t40 twice and eliminated quite a few that you would have caught with t45.

I think there were 2 that you split from the first few sequences before the two 8's that I observed but I had the impression that you took those 2 to t50 or t47 or something so I didn't include them. If you did, that would make 18. Now you've got even one more here, which would make 19.

When I posted the second group of 8 that had split, there were 145 sequences remaining. It seemed a little more than dwindling slightly. :-) At this point, knocking 10% off the list is not easy.

Look at it this way, could you run NFS and knock off 16 or 18 or 19 sequences in that amount of time? I'm thinking not even if you did only 150s and 151s. Here you're doing all kinds of huge sizes. It definitely feels like it's been worth it.

[EdH]

The confusion was that I was thinking back down the size of the cofactors: 189, 186, 184..., rather than down the sequences in the list.

A rough calculation says I should be able to GNFS a 180 digit composite in two weeks (which actually sounds amazing to me ATM). I guess three hours of ECM is a pretty small sacrifice for that. It's just that after several of those three hour failures, it becomes questionable. And, the ECM is a gamble while GNFS is going to work (most probably).