The NFS@Home automatic work generation infrastructure for "lasieved" 14e, inherited and improved from RSALS, forbids switching a number in postprocessing state back to sieving state (so as to prevent mistakes).
But once in a long while, the database needs an easy SQL update statement to perform that very manipulation ^^

[QUOTE=frmky;311935]That was undersieved! But I am surprised how relatively little difference 100M relations had on the matrix size.[/QUOTE]

Could be it's much denser. [URL="http://www.mersenneforum.org/showthread.php?p=311885#post311885"]I had a 1M matrix that took 25 minutes, but a 0.8M matrix that took 50 minutes.[/URL]

As long as Xyzzy is happy letting it run, so am I. :smile:

Some questions:

How are sieving results checked to make sure they are correct?

Is there a way to add bzip2 support to msieve?

[QUOTE=Xyzzy;312028]
How are sieving results checked to make sure they are correct?
[/QUOTE]

The same way a purported factor of a number is tested. ("Relation" is fancy for a number and its factors -- [or perhaps a few numbers and their factors. Regardless,] the information in a relation is factors.)

An observation:

We got bored today and started playing around with the number of cores allocated to each job. We got some interesting results. We assume that the results are from memory bandwidth limitations and/or "turbo" kicking in. (Because the "turbo" function boosts speed more when fewer cores are maxed out.)

Note that all systems are running 64-bit Debian from a SSD and that there is no swap partition or file at all. No other resource intensive programs are running. (If there were, it would mess up the timings.)

Obviously some jobs are further along than other jobs, but we think (?) the work is linear.

[CODE]╔═══════╤═════════════════════════╤═════════════════════════╤═════════════════════════╗
║ CORES │ i7-3770: S3m637 │ i5-2500: 8647_61_minus1 │ i5-2500: np_111 ║
╟───────┼─────────────────────────┼─────────────────────────┼─────────────────────────╢
║ 4 │ 124h00m │ 523h26m │ 151h03m ║
╟───────┼─────────────────────────┼─────────────────────────┼─────────────────────────╢
║ 3 │ * 114h32m * │ 468h43m │ 129h23m ║
╟───────┼─────────────────────────┼─────────────────────────┼─────────────────────────╢
║ 2 │ 130h39m │ * 413h04m * │ * 126h05m * ║
╟───────┼─────────────────────────┼─────────────────────────┼─────────────────────────╢
║ 1 │ 231h25m │ 557h15m │ 203h49m ║
╚═══════╧═════════════════════════╧═════════════════════════╧═════════════════════════╝[/CODE][SIZE=1]Snake1: Bow down and worship our most excellent ASCII box art![/SIZE]

Current post processing status…
 

[CODE]S3m637: linear algebra completed 6845180 of 14819292 dimensions (46.2%, ETA 113h55m)
8647_61_minus1: linear algebra completed 837523 of 16385983 dimensions (5.1%, ETA 412h28m)
np_111: linear algebra completed 54398 of 10242003 dimensions (0.5%, ETA 132h 7m)[/CODE]Note the speed increase from yesterday as a result of using fewer cores.

[QUOTE=Xyzzy;312030][SIZE=1]Snake1: Bow down and worship our most excellent ASCII box art![/SIZE][/QUOTE]
[URL="http://en.wikipedia.org/wiki/Emoticon#Orz"]Orz[/URL]

PS - We know the timings and status for np_111 look a little odd. We only tested each core selection for a few minutes so the timings may be a bit off. Over time, the timings become more accurate. (Look at the other two examples.)

The Core i5 is, in fact, a dual-core processor. 2 threads are better than 4 on the Core i5.
It's a bit odd than 3 threads are faster than 4 on the Core i7, which is a quad-core processor. It could indicate that you're limited by memory bandwidth, but I doubt your 16 GB of RAM are made of a single stick.

[QUOTE=debrouxl;312036]The Core i5 is, in fact, a dual-core processor. 2 threads are better than 4 on the Core i5.
It's a bit odd than 3 threads are faster than 4 on the Core i7, which is a quad-core processor. It could indicate that you're limited by memory bandwidth, but I doubt your 16 GB of RAM are made of a single stick.[/QUOTE]

Umm... [URL="http://ark.intel.com/products/52209/Intel-Core-i5-2500-Processor-(6M-Cache-up-to-3_70-GHz)"]i5-2500[/URL]* is a quad core with no hyperthreading.