Cunningham ECM efforts

[swellman]

Originally Posted by R.D. Silverman

You may want to consider holding off on the smaller numbers. The three you mention might be within
reach for NFS@Home. Greg needs to determine if he can do any of them.

I did communicate with Greg about the issue a few weeks ago. He encouraged me to move forward with ECM at the 25e9 level for all 23 remaining Cunninghams on the list. The first three will be enqueued tonight, after three erroneous values were removed. Mea culpa. I expect the first of these three to start ECM by Oct 1, the rest following every 4-6 weeks. If my estimate of Yoyo’s throughput is accurate(?), Greg/NFS@Home would eventually catch up to the ECM leading edge. But he has a few non-base-2 Cunningham jobs in his queue, which will take some months to process. So ECM of these first three Cunninghams absolutely should not delay Greg starting SNFS on the 23 survivors.

Further developments over this past weekend: @kruoli generously offered to run a subset of the remaining 23 Cunninghams through phase 1 of ECM at 25e9. He is limited by his GPU hardware to blocks of 5376 curves (ETA 12 days), but he has two such GPUs.

Yoyo is aware of all this and can run the phase 2 portion(s) on the phase 1 residues. This effort is still in the planning phase but we could be ready to move forward in October depending on current taskings. Throughput in earnest has yet to be determined of course, but at least a year could saved with this approach. I just hope a factor or two is found by ECM.

[xilman]

Originally Posted by swellman

Further developments over this past weekend: @kruoli generously offered to run a subset of the remaining 23 Cunninghams through phase 1 of ECM at 25e9. He is limited by his GPU hardware to blocks of 5376 curves (ETA 12 days), but he has two such GPUs.

Yoyo is aware of all this and can run the phase 2 portion(s) on the phase 1 residues. This effort is still in the planning phase but we could be ready to move forward in October depending on current taskings. Throughput in earnest has yet to be determined of course, but at least a year could saved with this approach. I just hope a factor or two is found by ECM.

I discovered the new facility to run stage-2 ECM at yoyo@home a couple of weeks ago when some strange looking entries appeared in the cw queue. On querying what was going on, I was told that Oliver had scheduled them and, indeed, they were in multiples of 5376 curves for each candidate.

No factors have been found but the reporting display seems a bit odd. All appear five appear to have completed all their curves but the display is still a solid red 0%. Perhaps Oliver may wish to take this up with Yoyo. However, that seems only to be a reporting issue because my traditional ECM jobs are being issued as normal.

[swellman]

Yoyo’s Phase 2 graphics have always been problematic. It’s a known issue. The graphic always reads “0%” but the curve count steadily increases over time. Some WU issued for Phase 2 are not returned from the volunteer worker, and end up being considered abandoned. These abandoned WUs are eventually reissued by the server (and finished) of course, but the curve count for Phae 2 never decrements. As a result, while Phase 1 may show 9900/9900 curves at 100% completed, Phase 2 often shows over 10000 curves completed. But they all get processed eventually.

ETA: The option of uploading a file called p2 with the residues from Phase 1 ECM into Yoyo’s work queue is a recent development. I’ve not tried it yet. Yoyo did warn me to drop him an email if I want to try it, as the process is not yet automagic, i.e. it takes manual intervention to start the process.