Trying the evalBuffer strategy

[jmct]

I've been talking with @reiddraper about the performance issues you ran into when trying to parallelise LSC.

He mentioned that while parMap may not have been suitable, perhaps a happy medium (not as much speedup in the good cases but not the very bad slowdown in the bad) could be reached with evalBuffer which allows you to chunk a list 'n' at a time for parallel reduction.

Any thoughts?

P.S. I wanted to link to the relevant line and be hip but since you've written much of your code in literate Haskell I can't do that. That seems to be a flaw in the way github deals with line numbers, and a flaw with you for being different and using .lhs ;)

[jasonreich]

Could be worth trying. I'll run some benchmarks. 

—
Sent from Mailbox for iPad

On Tue, Nov 5, 2013 at 6:50 PM, Jose Calderon [email protected]
wrote:

I've been talking with @reiddraper about the performance issues you ran into when trying to parallelise LSC.
He mentioned that while parMap may not have been suitable, perhaps a happy medium (not as much speedup in the good cases but not the very bad slowdown in the bad) could be reached with evalBuffer which allows you to chunk a list 'n' at a time for parallel reduction.

Any thoughts?

Reply to this email directly or view it on GitHub:
#10

[jmct]

How fast are you?!

Color me impressed.