Parallel-Haskell in industry
sebastien lannez
I have been playing with the Remote package from Hackage and it seems
to work pretty well.
The examples shipped with the code are very easy to adapt.
In order to set up an agent based system, I am currently looking at
environment to develop distributed applications.
The two most advanced systems (ie lightweight, functional programming,
easy to adapt, smart communication, ...) currently targeted are
Haskell.Remote and Erlang.
Before going further in the exploitation of Haskell.Remote, I would
like to know more about
1) performance limitation (in terms of number of computer nodes/
threads/sparks...) ?
2) communication needs (communication overhead -> low/high bandwith
requirement) ?
3) code stability (crash, memory leak,...) ?
4) already developed applications ?
briefly speaking I would like to know if it is stable for industrial
uses _or_ if it will suite industrial standard in the future.
Thanks,
------------------------------------------
Sébastien Lannez
sebastie...@gmail.com
http://www.lannez.fr
------------------------------------------
Jeff Epstein
I'm glad to hear you're interested in Cloud Haskell. I think it could
be a good platform to develop distributed applications. At the moment,
however, it is definitely research software, which means that while we
encourage you to use it, it is a work in progress. It certainly has
bugs, as-yet undiscovered performance peculiarities, scaling
limitations, and so on. In addition, as it's developed, its API may
change. In other words, don't stake your company on CH (at least not
yet).
If you do experiment with CH, please let us know about your results,
so we can improve it. Also, if you let us know what kind of
distributed applications you'd like to develop, maybe we can advise
you as to CH's suitability for the task.
Jeff
Duncan Coutts
> Dear all,
Hi Sébastien,
> I have been playing with the Remote package from Hackage and it seems
> to work pretty well.
> The examples shipped with the code are very easy to adapt.
Great.
[..]
> briefly speaking I would like to know if it is stable for industrial
> uses _or_ if it will suite industrial standard in the future.
As Jeff said, it's still slightly early days, though we're all pretty
happy with the design and we think it will be possible to make a robust
implementation with decent performance. This is mainly because the
design allows so much to be implemented as an ordinary Haskell library,
using the ordinary Haskell binary serialisation and network libraries,
rather than requiring tricky extensions to the runtime system.
I see you already contacted Simon. My company is currently running a
two-year project [1] for Simon to push commercial use of parallel
Haskell and we recently took on two partner companies that are very keen
on the cloud Haskell / remote stuff. So we are working on the
implementation and should be able to gather data to better answer
questions of performance, overheads etc.
That said, I wouldn't want to discourage you from trying it out and
sharing your experience. Certainly we would find that valuable.
[1]: http://www.haskell.org/haskellwiki/Parallel_GHC_Project
--
Duncan Coutts, Haskell Consultant
Well-Typed LLP, http://www.well-typed.com/