G197

[Andi47]

Originally Posted by chris2be8

That link doesn't work for me.

fivemack: fixed now

I don't know how others see this, but IMHO editing someone's posting like this has a disadvantage: as long as no new post (e.g. by some third person) appears in the thread, one doesn't recognize the edit by just looking into the subforum.

I once spent almost two days waiting for an answer after I posted some issue (I don't remember what exactly it was) - I repeatedly clicked into the subforum, saw "no new post in the respective thread", and therefore thought "no use clicking into the thread" - until I found out that an answer has been edited into my posting into my posting just shortly after I posted the question.

[fivemack]

After about a month running 48 threads at low priority on the 48-core machine, I've done the 80.0 - 89.6M region.

About 24000000 relations

Reserving 89.6M - 100M

[bdodson]

Originally Posted by bdodson

I'll start with 200M-300M, while our disks are crowded with data for the matrices
on two completed sieving tasks. -Bruce

Off to a slow start here, but these are reduced to the last few stragglers.
I'll take 300M-350M. -Bruce

[bdodson]

Originally Posted by bdodson

Off to a slow start here, but these are reduced to the last few stragglers.
I'll take 300M-350M. -Bruce

At last, the stragglers are all in, 200M-340M is done, and the last
of 340M-350M is queued. I'll take the rest of the last, 350M-400M,
which gets me somewhat past half of the sieving range. -Bruce

(Just 100M-200M left open.)

[fivemack]

You've probably already started 350-400, and that's not a problem, but I suspect we're not going to get another contributor and for pure tidiness it might be worth starting to work backwards from 200.

Would you mind trying an upload to ftp ssh.fivemack.org directory 7+374.sieve ? There's only 23GB free on that partition at the moment so it might not be enough space for everything you've done so far, but don't worry if you fill it.

I'm only up to 95.589 at the moment - running on the idle time of 48 cores which aren't all that idle.

[bdodson]

Originally Posted by fivemack

You've probably already started 350-400, and that's not a problem, but I suspect we're not going to get another contributor and for pure tidiness it might be worth starting to work backwards from 200.

Would you mind trying an upload to ftp ssh.fivemack.org directory 7+374.sieve ? There's only 23GB free on that partition

OK, I'll start by sending 200M-300M, which is 15.376 Gb. If we're not
expecting additional contributors I might prefer to follow 350M-400M
with 100M-150M, or perhaps 100M-170M, depending upon how much
of 100M-200M I'd be running. Ah, looks like 100M-180M or maybe
100M-190M. That is to say, I'd be happier working up (rather than
backwards!), and I'd prefer to avoid 190M-200M.

-Bruce

at the moment so it might not be enough space for everything you've done so far, but don't worry if you fill it.

I'm only up to 95.589 at the moment - running on the idle time of 48 cores which aren't all that idle.

[fivemack]

With 80-95 + 200-300:

Thu Sep 15 20:04:54 2011 found 39492312 hash collisions in 284216780 relations
Thu Sep 15 20:46:33 2011 found 29763638 duplicates and 254574782 unique relations
Thu Sep 15 22:57:19 2011 begin with 254574782 relations and 355034161 unique ideals

With 80-95 + 200-350:

Fri Sep 16 03:17:37 2011 found 68933027 hash collisions in 398349772 relations
Fri Sep 16 04:22:00 2011 found 53607920 duplicates and 344741852 unique relations
Fri Sep 16 07:21:04 2011 begin with 344741852 relations and 401907781 unique ideals

So 114 million lines got 90.1 million unique relations and 46.8 million extra ideals; the gap is closing well, the next 50MQ should get to an incredibly ugly and huge matrix, and the 50MQ after that ought to make the matrix tractable.

[bdodson]

Originally Posted by fivemack

With 80-95 + 200-300:
...
With 80-95 + 200-350:
...
So 114 million lines got 90.1 million unique relations and 46.8 million extra ideals; the gap is closing well, the next 50MQ should get to an incredibly ugly and huge matrix, and the 50MQ after that ought to make the matrix tractable.

In that case, I'll reserve 100M-170M and plan on running that once 350M-400M
finishes. -Bruce

(Leaving 170M-200M open if there's anyone that still wants a sample of
the sieving, now that we can see the prospect of g197 sieving finishing
on the horizon.)

[fivemack]

80-100 now complete; reserving 190-200, this ought to take less than the six weeks for 89.6-100 because I won't be running fifty aliquot sequences at the same time.

[fivemack]

With 220MQ (80-100 + 200-400), I got

Thu Sep 29 12:43:36 2011  found 105904252 hash collisions in 520302381 relations
Thu Sep 29 18:06:12 2011  begin with 434853690 relations and 434086720 unique ideals
Thu Sep 29 18:47:02 2011  reduce to 194156130 relations and 160378269 ideals in 28 passes
Thu Sep 29 21:06:31 2011  memory use: 5512.0 MB
Thu Sep 29 21:06:32 2011  reading all ideals from disk
Thu Sep 29 21:07:20 2011  memory use: 8650.7 MB
Thu Sep 29 21:09:18 2011  keeping 201969981 ideals with weight <= 200, target excess is 1017883
Thu Sep 29 21:11:32 2011  commencing in-memory singleton removal
Thu Sep 29 21:12:50 2011  begin with 194156134 relations and 201969981 unique ideals
Thu Sep 29 21:47:37 2011  reduce to 193215727 relations and 201029094 ideals in 23 passes
Thu Sep 29 21:47:37 2011  max relations containing the same ideal: 200
Thu Sep 29 21:49:06 2011  filtering wants 1000000 more relations

So just over half a billion relations isn't quite enough for this 33LP case, but it's close; when 100M-200M is done I expect the matrix to be not completely horrendous.

[fivemack]

80-150 + 200-400

Fri Oct  7 08:51:01 2011  found 156746781 hash collisions in 651410238 relations
Fri Oct  7 08:53:23 2011  commencing duplicate removal, pass 2
Fri Oct  7 11:54:52 2011  found 138488186 duplicates and 512922052 unique relations
Fri Oct  7 17:25:58 2011  begin with 512922052 relations and 458399228 unique ideals
Fri Oct  7 18:04:53 2011  reduce to 317129864 relations and 242177207 ideals in 16 passes
Sat Oct  8 05:36:02 2011  commencing 2-way merge
Sat Oct  8 05:41:39 2011  reduce to 94741713 relation sets and 92788558 unique ideals
Sat Oct  8 05:41:39 2011  commencing full merge
Sat Oct  8 06:40:43 2011  memory use: 10489.0 MB
Sat Oct  8 06:41:09 2011  found 51833810 cycles, need 51564758
Sat Oct  8 06:41:59 2011  weight of 51564758 cycles is about 3609615537 (70.00/cycle)

Still sieving (and of course need to do the next filter with a higher trial density)