On Wed, Jan 13, 2010 at 10:54 AM, Dr. David Kirkby <david.kir...@onetel.net> wrote: > William Stein wrote: >> >> On Wed, Jan 13, 2010 at 12:42 AM, Dr. David Kirkby >> <david.kir...@onetel.net> wrote: >>> >>> Can someone give me a Sage command to type that can exploit multiple >>> processors, which I can then time? (I'm not a Sage user really, just >>> helping >>> port it to Solaris, so don't rely on me knowing python). >>> >>> Ideally a floating point and an integer example would be good. >> >> Here is an example: >> >> sage: time b = bernoulli(10^5, algorithm='bernmm', num_threads=24) >> CPU times: user 5.61 s, sys: 0.09 s, total: 5.70 s >> Wall time: 0.82 s >> >> Just to emphasize -- it's wall time that matters. > > > Em, best I could manage on t2 (in the global zone, so no CPU capping) was > 24.25 s with 256 threads. That's twice the number of threads that t2 has > (128). > > That makes it 31 times slower than whatever you used!
I used boxen.math, which is a 24 core Xeon 2.6Ghz. What matters for this benchmark is the number of cores that the computer has. Though t2 can manage 128 hardware threads, it only has 16 actual *cores*. We have on t2: sage: time b = bernoulli(10^5, algorithm='bernmm', num_threads=1) CPU times: user 118.64 s, sys: 0.01 s, total: 118.64 s Wall time: 118.77 s and on boxen: sage: time b = bernoulli(10^5, algorithm='bernmm', num_threads=1) CPU times: user 5.40 s, sys: 0.00 s, total: 5.40 s Wall time: 5.42 s so individual cores on boxen are 22 times faster than individual cores on boxen, for this computation. We have sage: 24*22/16.0 33.0000000000000 which confirms what you observed. My understanding is that though the T2 processor can very quickly switch between hardware threads, it still only has 16 cores. There probably exist programs (web servers, *virtualization*, etc.) such that this thread switching is a major bottleneck. It seems it is irrelevant to probably most math and scientific computation. -- William > > > In the zone on > > http://t2nb.math.washington.edu:8000/ > > times are slightly longer, as the CPU usage is capped a bit. > > Dave > > kir...@t2:[~/sage-4.3] $ uname -a > SunOS t2 5.10 Generic_141414-02 sun4v sparc SUNW,T5240 > kir...@t2:[~/sage-4.3] $ ./sage > ---------------------------------------------------------------------- > | Sage Version 4.3-patched-for-Solaris(SPARC), Built 2009-12-26 | > | Type notebook() for the GUI, and license() for information. | > ---------------------------------------------------------------------- > sage: time b = bernoulli(10^5, algorithm='bernmm', num_threads=1024) > CPU times: user 226.64 s, sys: 36.21 s, total: 262.86 s > Wall time: 24.56 s > sage: time b = bernoulli(10^5, algorithm='bernmm', num_threads=512) > CPU times: user 225.75 s, sys: 38.95 s, total: 264.70 s > Wall time: 24.54 s > sage: time b = bernoulli(10^5, algorithm='bernmm', num_threads=256) > CPU times: user 223.24 s, sys: 16.38 s, total: 239.62 s > Wall time: 24.25 s > sage: time b = bernoulli(10^5, algorithm='bernmm', num_threads=128) > CPU times: user 219.54 s, sys: 6.95 s, total: 226.49 s > Wall time: 24.55 s > sage: time b = bernoulli(10^5, algorithm='bernmm', num_threads=64) > CPU times: user 160.53 s, sys: 1.34 s, total: 161.87 s > Wall time: 24.53 s > sage: time b = bernoulli(10^5, algorithm='bernmm', num_threads=32) > CPU times: user 130.78 s, sys: 0.14 s, total: 130.93 s > Wall time: 25.37 s > sage: time b = bernoulli(10^5, algorithm='bernmm', num_threads=16) > CPU times: user 121.34 s, sys: 0.07 s, total: 121.41 s > Wall time: 27.56 s > sage: time b = bernoulli(10^5, algorithm='bernmm', num_threads=8) > CPU times: user 119.47 s, sys: 0.03 s, total: 119.50 s > Wall time: 32.82 s > sage: time b = bernoulli(10^5, algorithm='bernmm', num_threads=4) > CPU times: user 119.28 s, sys: 0.03 s, total: 119.32 s > Wall time: 44.08 s > sage: time b = bernoulli(10^5, algorithm='bernmm', num_threads=2) > CPU times: user 119.16 s, sys: 0.02 s, total: 119.18 s > Wall time: 68.00 s > sage: time b = bernoulli(10^5, algorithm='bernmm', num_threads=1) > CPU times: user 118.64 s, sys: 0.01 s, total: 118.64 s > Wall time: 118.77 s > > > > > -- > To post to this group, send an email to sage-devel@googlegroups.com > To unsubscribe from this group, send an email to > sage-devel+unsubscr...@googlegroups.com > For more options, visit this group at > http://groups.google.com/group/sage-devel > URL: http://www.sagemath.org > > -- William Stein Associate Professor of Mathematics University of Washington http://wstein.org
-- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
