[Dailydave] Parallelism

Kristian Erik Hermansen kristian.hermansen at gmail.com
Mon Oct 29 15:24:20 EDT 2007


On 10/29/07, Dave Aitel <dave at immunityinc.com> wrote:
> Does anyone else find it odd that Elcomsoft is getting a lot of press
> for something relating to cracking on video cards, when you could just
> write your cracker in RapidMind and be done with the whole problem? No
> doubt that was one of RapidMind's early test programs.

CPUshare global computing grid + JohnTheRipper == sexy
https://www.cpushare.com/wiki/cpushare/JohnTheRipper

> In any case, it's interesting that Linux Journal and MSDN Magazine are
> both focused on parallelism lately. Both of them are a bit "off" in my
> opinion. The answer is going to be a language that abstracts data
> access that extra layer to make parallelism something your compiler
> does for you. It's not going to be parallel additions to already
> non-parallel languages. You have to love the "caveats" in MSDN
> Magazine about using "Only one parallel loop at a time". And also
> "Parallel exceptions in LINQ are non-deterministic"! Imagine the
> hilarious security side effects from these sorts of half-backed
> parallel solutions.

You are right.  Most developers can't even prevent simple bugs in
their programs, let alone think about parallelism as well.  What I
find more interesting is that people always love their faster
multi-core CPUs, but let's say you have a Quad Core with a
single-threaded app ... "oh damn!!!  I just increased the running time
of this app by 4x over a single core CPU".  So, in general, multi-core
CPUs actually make a lot of code 'less-efficient'.  I believe that the
Intel compiler (icc) has some of this optimization for multi-core
built in, but it has not trickled down to the open source compilers
like gcc/g++ yet.  Even still, it is not a simple problem to solve,
and in fact, this was a topic of discussion for some time at various
points in history going back at least 15 years.  When I spoke at the
Ubuntu Live conference last summer, there was a guy from Intel there
talking about the problem and their new open source tool Threading
Building Blocks (TBB).  Links below...

http://arstechnica.com/journals/linux.ars/2007/08/03/ars-at-ubuntu-live-intels-tbb-announcement-party
http://www.ubuntulive.com/cs/ubuntu/view/e_sess/13467

The presentation file:
http://conferences.oreillynet.com/presentations/ubuntu2007/ul_reinders.pdf
-- 
Kristian Erik Hermansen


More information about the Dailydave mailing list