GNFS poly search time limit oddity - Page 2

akruppa · 2009-02-25, 00:30

Oh, that's right. You could try to call clock() every so often and add up elapsed seconds or so, but maybe it's easier to use getrusage() right away.

Alex

Brian Gladman · 2009-02-25, 08:13

The other issue that needs attention when the cycle counter is used for timing is that the counters on the different cores of x86 multi-core processors are not synchronised.

So, unless a process is forced onto the same processor core, two successive values cannot be used simply for interval timing.

Brian Gladman

akruppa · 2009-02-25, 12:33

I learned that the hard way recently. I had a real hard time getting reliable timings (using the cycle-counter) for our Schönhage-Strassen code, until finally the cpus in our dual Opteron nodes were desynchronized enough that I started getting negative timings. That gave it away. numactl to the rescue!

Alex

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Search result limit	davar55	Forum Feedback	15	2019-02-26 15:41
Poly search candidates	schickel	Msieve	32	2013-11-05 19:11
Poly Search vs Sieving times	EdH	Factoring	10	2013-10-14 20:00
GNFS poly selection	frmky	Factoring	14	2012-07-23 01:57
gpu poly search error	bdodson	Msieve	10	2010-11-09 19:46

2009-02-25, 00:30	#12
akruppa "Nancy" Aug 2002 Alexandria 2,467 Posts	Oh, that's right. You could try to call clock() every so often and add up elapsed seconds or so, but maybe it's easier to use getrusage() right away. Alex

2009-02-25, 08:13	#13
Brian Gladman May 2008 Worcester, United Kingdom 1024₈ Posts	The other issue that needs attention when the cycle counter is used for timing is that the counters on the different cores of x86 multi-core processors are not synchronised. So, unless a process is forced onto the same processor core, two successive values cannot be used simply for interval timing. Brian Gladman

2009-02-25, 12:33	#14
akruppa "Nancy" Aug 2002 Alexandria 2,467 Posts	I learned that the hard way recently. I had a real hard time getting reliable timings (using the cycle-counter) for our Schönhage-Strassen code, until finally the cpus in our dual Opteron nodes were desynchronized enough that I started getting negative timings. That gave it away. numactl to the rescue! Alex