Aliqueit.exe discussion

[Andi47]

Originally Posted by mklasson

Awesome! That's with an optimised 64-bit gnfs-lasieve* under linux though, right? I wonder what the optimal cutoff is for a sluggish windows system...

it's a 32-bit gnfs-lasieve* running under a sluggish windows XP system.

I have seen a benchmark over there in the msieve-1.42 beta feedback thread, saying that quartics sieve ~40% faster than quintics on a c95. This figure might be even better for a c92.

[mklasson]

Originally Posted by Andi47

it's a 32-bit gnfs-lasieve* running under a sluggish windows XP system.

I have seen a benchmark over there in the msieve-1.42 beta feedback thread, saying that quartics sieve ~40% faster than quintics on a c95. This figure might be even better for a c92.

Wow. Sounds like it's high time I tried out the new msieve.

[henryzz]

i will run some 64 bit gnfs-lasieve* benchmarks tomorrow
i will be comparing with 64-bit msieve instead of yafu
i will do some small ones like a c85 and maybe even c80


BTW what is the smallest nfs people think has ever been done?

[jasonp]

Originally Posted by henryzz

BTW what is the smallest nfs people think has ever been done?

The first SNFS factorization was 2^128+1, back in the mid-1980s. It took about 3 hours.

Introductory NFS papers regularly use NFS to factor 11-digit numbers :)

[henryzz]

Originally Posted by Andi47

I just benchmarked GNFS with a quartic against yafu on a c92: GNFS takes a total time of 01:21:20 hours, yafu takes 01:50:00 hours, so GNFS is quite a bit faster on this input size.

So for all who are already using msieve 1.42 beta, I suggest lowering the GNFS_cutoff in the aliqueit.ini file to 90 digits (BTW: If I understood correctly, 90 digits is the minimum input size where msieve will do GNFS poly selection.).


Edit: Does wget need to be in the aliqueit directory if I want to automatically upload new lines of my aliquot sequences to the database?

i presume you included poly selection time in your gnfs time

[Andi47]

Originally Posted by henryzz

i presume you included poly selection time in your gnfs time

Yes, I did, as I described in the msieve feedback thread linked in my previous post.

BTW: I don't think that msieve will do c85 poly selection - according to Jasonp, the lower limit for this is "slightly below 90 digits".

Nevertheless it would be interesting to benchmark some c8x's (with a "hacked" msieve, or possibly (feature request -->) a msieve 1.42 beta2 which hopefully will allow to GNFS c8x's.)

[henryzz]

Originally Posted by Andi47

Yes, I did, as I described in the msieve feedback thread linked in my previous post.

BTW: I don't think that msieve will do c85 poly selection - according to Jasonp, the lower limit for this is "slightly below 90 digits".

Nevertheless it would be interesting to benchmark some c8x's (with a "hacked" msieve, or possibly (feature request -->) a msieve 1.42 beta2 which hopefully will allow to GNFS c8x's.)

as soon i i had compiled it i did that hack and then recompiled
i virtually always do
i am currently running a c85 benchmark and had a bit of trouble because there are no good parameters for <c90 and as such the polynomial selection would have taken ages and messed up the benchmark
in the end i decided to time 2 minutes on my watch and press ctrl-c to stop the polynomial selection
it then ran for another 2 minutes and then factmsieve.pl was started by aliqueit
the only downside to that method is that it is quite possible that the polynomial would have been improved on with better parameters or more time
edit:
another possiblity to consider is:
when will gnfs-lasieve4I11e become better?

[henryzz]

c85 64-bit asm ggnfs with msieve poly selection and postprocessing
real 25m44.344s
user 24m21.527s
sys 0m10.333s

this makes me think the 4 minutes of poly selection was too much

msieve 64-bit qs
real 17m3.005s
user 16m49.299s
sys 0m1.156s

[henryzz]

i will test the same c85 on a 32-bit operating system tomorrow

[Andi47]

Originally Posted by mklasson

Either there or somewhere included in your PATH env. var. Am I the only one still actively using the PATH variable? A nice directory such as c:\utils with all sorts of useful tools in it is very handy imho.

I intended to put it into the aliqueit folder and got a "wget not found" error. Seems that aliqueit.ini needs a "wget_path=..." variable...

Edit: found that the binary and the .dll files have been dropped to a .bin folder, now it seems to work. But....

After putting the factors to the database, aliqueit doesn't do further work on the sequence. It would be nice if it did, so that it could drop new lines to the database as soon as they are found.

[mklasson]

Originally Posted by Andi47

After putting the factors to the database, aliqueit doesn't do further work on the sequence. It would be nice if it did, so that it could drop new lines to the database as soon as they are found.

Yeah, but we wouldn't want to halt the factoring progress while waiting for the connection/transfer to complete. I guess I could start wget from another thread though. My current setup is to let task scheduler run a submit on all my sequences every 6 hours, but yeah, it would be nicer to let the computing aliqueits take care of the submitting as well.