Re: NG Lush and concurrency

On Thu, 30 Mar 2006, Yann LeCun wrote:

> For one thing, the low-level numerical functions could all
> be implemented on top of an optimized library that can take
> advantage of multiple cores/processors (e.g. Intel mkl,
> Atlas,....).

yes, that would be nice to have today already. MKL is not
free though.


> For the higher level, we could provide parallel IDX
> iterators. There is already idx-pbloop which is supposed
> to be a parallel version of idx-bloop (but currently just
> calls idx-bloop).

something like that. how would you implement them, though?
spawn a thread for every slice (if the slices are big enough)?
...


> For higher-level granularity, we could use MPI, and for
> coarse-grained parallelism, we can run multiple Lush processes
> through the remotelush class.

i have no experience with those. aren't they in the same
department in terms of granularity (I think of MPI as
infrastructure for computing on clusters)?

ralf




-------------------------------------------------------
This SF.Net email is sponsored by xPML, a groundbreaking scripting language
that extends applications into web and mobile media. Attend the live webcast
and join the prime developer group breaking into this new coding territory!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642