Just a stupid side-question:
On Sun, 4 Oct 2015, Victor Shoup wrote:
Also, now that NTL is threadsafe, I'm looking at making some of the
low-level routines thread enhanced.
For up to 4 threads, I get close to linear speedup...but after that, it
starts to degrade: at 16 threads I only get 8x speedup.
Isn't this what always happen with parallel computing? Computing in 1
thread takes for example 100 seconds. With 2 cores it takes 100/2+1=51,
assuming that overhead for thread handling is 1 second. And at some point
overhead will be too much, or memory IO will be bottleneck.
Now let's suppose that Sage has users A and B. A wants to compute one
hard thing - say, factor one big integer. B wants to compute many things -
say, factor 100 medium-size integers. For B it would be best to have no
threading at all on lower level, just very simple upper level wrapper.
I have been asked to answer mostly type B questions, but sometimes also of
type A.
* * *
If I am right, there is no EASY parallel functions in Sage. Let G be a
list of 1000 immutable graphs. How to run
C = [g.chromatic_number() for g in G]
in parallel? AFAIK we have no "parallel_apply()" or similar.
--
Jori Mäntysalo
