mersenneforum.org Workunit diet ?
 Register FAQ Search Today's Posts Mark Forums Read

 2004-02-15, 03:11 #1 dsouza123     Sep 2002 2×331 Posts Workunit diet ? What has happened with the status of 811M ? It was supposed to be about 80 000 workunits and was a little over 50% complete, now it needs 61 000 workunits and is about 71% complete. The stats on the webpage have stopped at Feb 12.
 2004-02-16, 14:56 #2 junky     Jan 2004 100001012 Posts i've remark the total WUs have dropped from 78 000 to 61 000 too. I've a question related to that number, do we need to complete all these WUs (total) to complete the 811M ? I mean, could we solve that contest at like 90% like we can do with ECC2-109 or RC5-72 ? if we say, we're doing 2%/day, so in ~ 13 days, that contest will be over. which one is plan to be done ? Just to take a look at that new one. Thanks.
 2004-02-16, 16:01 #3 dsouza123     Sep 2002 66210 Posts As I understand it, it is really the relations that are the issue. Calculations are done before starting M811 to determine/estimate how many relations are needed for a successful run. When the linesieving is finished other steps are then done which end up with the factorization. If not enough relations were gathered it may not be possible to complete the factorization. I believe the NFSNET team running NFSNET have the mathematical understanding and sufficient experience to determine the number of relations needed. The workunits are project independent units of work that allow comparing work done in this project with other previous ones. Maybe the scale of workunits to relations has been recalculated. The ratio of (61432931 relations done) / (82961419 around the inital estimate) = 74.05 % which is the current completion percentage. So even with the number of workunits reduced by about 1/4 the completion ratio still appears correct.
2004-02-17, 09:43   #4
xilman
Bamboozled!

"ð’‰ºð’ŒŒð’‡·ð’†·ð’€­"
May 2003
Down not across

254178 Posts

Quote:
 Originally Posted by dsouza123 As I understand it, it is really the relations that are the issue. Calculations are done before starting M811 to determine/estimate how many relations are needed for a successful run. When the linesieving is finished other steps are then done which end up with the factorization. If not enough relations were gathered it may not be possible to complete the factorization. I believe the NFSNET team running NFSNET have the mathematical understanding and sufficient experience to determine the number of relations needed. The workunits are project independent units of work that allow comparing work done in this project with other previous ones. Maybe the scale of workunits to relations has been recalculated. The ratio of (61432931 relations done) / (82961419 around the inital estimate) = 74.05 % which is the current completion percentage. So even with the number of workunits reduced by about 1/4 the completion ratio still appears correct.
I posted a brief explanation of this phenomenon to the nfsnet-l mailing list but failed to include one here. My apologies.

Two things have happened. The first is that we made a modest adjustment downwards in the number of work units required to complete the computation. This adjustment took the total down from 80k WU to78k WU and was done because the amount of work completed to that point enabled us to make a much more precise estimate of the total amount that would be required. The initial estimate was based on perhaps 1 ten-thousandth of the total required and although the estimates are pretty precise they can't be perfect except by chance.

The major change to the estimate arises from an entirely different cause. Three of us are experimenting with an off-line siever which has turned out to be quite productive, as can be seen from the recent increase in relations sent in at http://www.nfsnet.org/stats2/statsre...ate=index.html

The increased productivity means that a smaller amount of work must be done by the standard NFSNET clients, which is reflected in the lower estimate for the total WU required.

When we started the experiment we weren't sure whether it would even work at all, so we didn't make any prior announcement. It is indeed working, but it's very far from being ready for prime time. The process needs a great deal of human attention and until it is automated there is no way we could support it generally. Later this year, perhaps.

Paul

 2004-02-17, 19:42 #5 junky     Jan 2004 7·19 Posts that's great Paul, if we can reduce the time of work to complete the 811M
 2004-02-27, 00:42 #6 junky     Jan 2004 7×19 Posts i've remarked the number droped again. from 61 000 to 58 000 workunits. so we reached the number of 85% nice job to everyones.

 Similar Threads Thread Thread Starter Forum Replies Last Post Christenson Hardware 32 2011-12-25 08:17 garo Hardware 41 2011-10-06 04:06 Ungelovende Prime Sierpinski Project 8 2009-09-14 18:00

All times are UTC. The time now is 03:25.

Sun Nov 28 03:25:11 UTC 2021 up 127 days, 21:54, 0 users, load averages: 0.94, 0.95, 0.93

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.