Date: Sat, 19 Nov 2011 19:55:50 -0500 From: Stephen Reese <rsreese@...il.com> To: john-users@...ts.openwall.com Subject: Re: OpenMP not using all threads On Sat, Nov 19, 2011 at 5:58 PM, Solar Designer <solar@...nwall.com> wrote: > On Sat, Nov 19, 2011 at 04:29:12PM -0500, Stephen Reese wrote: >> I have patched john-1.7.8.tar.gz with john-1.7.8-omp-des-7.diff.gz in > > -omp-des-7 is good if you want to attack just one salt or very few > salts. For many salts, -omp-des-4 provides better performance. (This > is mentioned on the wiki.) > > Alternatively, if you feel adventurous, you may do a CVS checkout for > even newer code (currently known as development version 188.8.131.52), which > combines the best properties of these two patches into one source code > tree (no patches are needed). The CVS checkout instructions are here: > > http://www.openwall.com/Owl/DOWNLOAD.shtml > >> order to utilize four threads from a E5520 on a Debian system but it > > Actually, you should be running 8 threads on this CPU unless you have > Hyperthreading disabled. But you don't need to worry about that - gcc's > libgomp will run as many threads as you have logical CPUs by default > (most likely 8). > >> instead seem like it is only using two. When testing DES I see around >> 2500K c/s and when patched about 5000K. I was hoping for closer to >> 8000K to 10000K. > > Yes, you should get about 9000K (for 8 threads combined). > > The increase may be less than 4x because the thread-safe code is slower, > because the CPU clock rate is lower when all cores are in use (E5520 has > Turbo Boost), and for certain other reasons. Yet you should in fact get > 9000K or so. > >> I also edited the Makefile as follows: >> >> # gcc with OpenMP >> OMPFLAGS = -fopenmp -msse2 >> >> Another strangeness is when testing is I am not seeing the -16 >> appending to the following: >> >> Benchmarking: Traditional DES [128/128 BS SSE2]... DONE >> >> Is this normal or did something go wrong. > > It looks like you made a 32-bit build. That is, you probably used the > linux-x86-sse2 make target instead of linux-x86-64. When you build > without OpenMP, the -sse2 target uses assembly code supplied with JtR, > however when you go for OpenMP, gcc has to generate thread-safe code > instead. It does this well for x86-64, but not for 32-bit x86 (there > are too few registers on 32-bit x86). > > To get decent performance at DES with OpenMP builds on your machine, you > ought to make 64-bit builds. And indeed your install of Debian should > be 64-bit, too. > > I hope this helps. > > Alexander > > P.S. The Subject is almost certainly wrong - there's no indication that > the build doesn't use all threads. Rather, the threads are slow. > Alexander, Thanks for the great information and noted about the Subject line. The tests were on a Linode which is shared XEN hosting. I had a feeling that the 32-bit architecture might be an issue as I noticed that "OpenMP example" was only twice as fast (32-bit OpenMP) instead of four times (64-bit OpenMP). http://openwall.info/wiki/internal/gcc-local-build#OpenMP-example. Though OpenMP example is four times as fast neither the CVS nor stable/patch versions of John would provide the 4x speed-up I was hoping for even on the 64-bit. Maybe XEN and the other respective hosts across the multiple Linodes I am testing are causing roughly a 45 - 60% slowdown from a bare-metal instance but not affecting the "OpenMP Example". root@:~# time ./loop2 615e5600 real 0m2.229s user 0m2.226s sys 0m0.002s root@:~# time ./loop 615e5600 real 0m0.333s user 0m1.313s sys 0m0.003s What I am trying to achieve: I have 42 DES passwords and three Linodes. Password list is currently split-up so each host has 12 entries and are running in incremental mode. Is there a better way, such as specifying a thread per instance on a single host? Is there a performance/time benefit in splitting up the password list amongst multiple hosts or is one host going to achieve the same results as the three?
Powered by blists - more mailing lists
Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.