New BOINC framework for factoring

[jasonp]

This codebase was announced by Tom Ritter at the EKO security conference, and was actually used to factor a 512-bit RSA key in two days.

Greg, is there anything that NFS@Home can use in here?

(the GNFS subdirectory has some nice statistical visualizations too)

PS: The slides mention a service that factors 512-bit RSA keys at EUR5000 each. He's probably talking about these guys.

[debrouxl]

Interesting

I'd say that he used about 50 full-time cores, because factoring a 512-bit RSA key takes at worst 100 CPU-days on recent quad-core desktop computers.

Originally Posted by debrouxl

I'd say that he used about 50 full-time cores, because factoring a 512-bit RSA key takes at worst 100 CPU-days on recent quad-core desktop computers.

100 CPU-days on a quadcore is 400 core-days, or 9600 core hours.
I had estimated, and seen, around 460 +/- instance-hours on a 8-core in Amazon (High-CPU Extra Large) - that's 3680 core-hours.
Not sure where the discrepancy is there.

The original cash-for-primes people were these people whose site is gone, but references still exist.

If you had problems downloading the notes pdf from github, it's a github issue.
You can get it here: http://ritter.vg/misc/stuff/Cloud%20...0-%20notes.pdf

[yoyo]

Hello,

don't know if you saw it yet, I run now YAFU as Boinc project: http://yafu.dyndns.org/yafu/. I use the original unmodified yafu and gnfs binaries together with the Boinc wrapper.
The project factors composites from factordb, beginning with C80.
The number of unfactored composites there was already reduced from 250k to 200k in the last weeks.

yoyo

[debrouxl]

Welcome to MersenneForum

My bad, I meant ~100 core-days, i.e. 2400 core-hours, which makes for a smaller discrepancy

Indeed, Benjamin "FloppusMaximus" Moody, in the United-TI posts that he removed after receiving TI's illegal DMCA takedown requests, estimated his solo factorization of the 512-bit RSA public key for TI-83+ OS signature checking to the equivalent of ~73 full-time calendar days, on a dual-core Athlon 64 computer.
So I estimated that newer micro-architectures, and the polynomials selected by msieve (which yield more relations per time unit than those generated by pol51, which Benjamin Moody, and we at RSALS, used in 2009), should reduce the amount of work around 100 core-days.