M991 reservations thread

[xilman]

Originally Posted by Gimarel

Please append M991.140-142.7z.part2 to M991.140-142.7z:

 cat M991.140-142.7z.part2 >> M991.140-142.7z

310312968 + 262 = 310313230

That worked fine. Thanks, and apologies for my stupidity. Unfortunately it's easy to become confused with dozens of files and tens of gigabytes of data.

I also found a readme which told me which archive file I should have used to solve another case.

The only remaining file which hasn't been processed is M991.394-396M.7z. It was uploaded on Feb 18th, apparently by Rich Dickerson. I'll contact him separately by email.

Results of analysis will be posted when I have them. Here's an extremely rough estimate based on the size of the files in total and the size of 1000 lines chosen at random.

anubis raw # ls -s | head -1
total 69386812
anubis raw # head -1000 191-192.out | wc
   1000    1000  124060
anubis raw # bc
bc 1.06.95
Copyright 1991-1994, 1997, 1998, 2000, 2004, 2006 Free Software Foundation, Inc.
This is free software with ABSOLUTELY NO WARRANTY.
For details type `warranty'. 
69386812 * 1024 / 124060 * 1000
572723000

We can conclude that there are over 570M relations but have no idea how many are duplicates.

[RichD]

Completed 378-382M. remdups4 says:

Found 8246331 unique, 15542 duplicate, and 12 bad relations.

[fivemack]

14702920643 characters
118206412 lines
114111349 unique relations

Taking 410-422

[xilman]

Rich sent in the missing 919 bytes and a further complete set of relations.

Just counting them all with wc took 34 minutes of cpu, so more details might not be published for a while. At the moment, there are 582334932 relations taking a total of 72604609829 bytes --- 582M and 72G if you don't want to count digits.

With Tom's report above, we've a total of 700541344 relations including all dups, 696446281 without Tom's dups, or 700M to adequate precision. Not entirely sure how many Tom will need to build a usable matrix. Total storage is 87307530472 bytes.

Paul

[fivemack]

Would it be reasonable to suggest the end of March for a first linear algebra attempt? I ought to have fifty million more relations locally by then.

[fivemack]

Reserving 422 to 438

[RichD]

I'll take 438-440M next.

[VBCurtis]

Originally Posted by xilman

With Tom's report above, we've a total of 700541344 relations including all dups, 696446281 without Tom's dups, or 700M to adequate precision. Not entirely sure how many Tom will need to build a usable matrix. Total storage is 87307530472 bytes.

Paul

According to Jason, NFS@Home's 3,697- GNFS-221 factorization used 33bit lp and 740M relations to build a density-112 matrix. Since this project is quite a bit easier, we should need fewer relations, right? Or, a similar number of relations should produce a smaller matrix? I guess that "usable" is important in the quote above. 650M might be enough for filtering to work, but the matrix would need more memory than we have available.

[fivemack]

382-394 done.

24518079 relations (24385140 unique), 7160 thread-hours (3800 on i7/4770K, 3360 on i7/4790K)

(updating this post after re-running the sections where I lost data in a disc-full incident)

[Gimarel]

85M - 96M is done: 15843215 unique and 134353 duplicate relations.

[Dubslow]

bill@Guilty-Spark:~/ggnfs$  ./remdups4 M991.396-400 
Counting rels in M991.396-400. This might take a few minutes for large rel files.
Found 8111523 unique, 14514 duplicate (0.2% of total), and 0 bad relations.
Largest dimension used: 51 of 80
Average dimension used: 24.8 of 80
bill@Guilty-Spark:~/ggnfs$ time bzip2 M991.396-400 

real    1m28.836s
user    1m22.673s
sys     0m0.603s
bill@Guilty-Spark:~/ggnfs$ wc -c M991.396-400.bz2 
486913308 M991.396-400.bz2

Available for download at http://dubslow.tk/random/M991.396-400.bz2