Parallel + exceptions

Andrew Coppin

23 Jun 2007 23 Jun '07

1:59 p.m.

The two things mentioned in the subject line are both things I've never tried with Haskell. I've seen a lot of papers about these things, but I don't really understand what the current state of play with this is. Are any of these things "production-ready" yet? And how do you use them in the real world?

Show replies by date

Neil Mitchell

23 Jun 23 Jun

2:21 p.m.

Hi Parallel: I think this is pretty well solved with STM. I think STM is production ready in GHC. Exceptions: These are less well solved, people are still thinking about some of the issues. In general, you probably don't want to use exceptions too much - the Maybe Monad is often a better choice. Thanks Neil On 6/23/07, Andrew Coppin wrote:

...

The two things mentioned in the subject line are both things I've never tried with Haskell. I've seen a lot of papers about these things, but I don't really understand what the current state of play with this is. Are any of these things "production-ready" yet? And how do you use them in the real world?

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

Bulat Ziganshin

2:52 p.m.

Hello Andrew, Saturday, June 23, 2007, 5:59:45 PM, you wrote:

...

The two things mentioned in the subject line are both things I've never tried with Haskell. I've seen a lot of papers about these things, but I don't really understand what the current state of play with this is. Are any of these things "production-ready" yet? And how do you use them in the real world?

definitive reading: "Tackling the awkward squad: monadic input/output, concurrency, exceptions, and foreign-language calls in Haskell" http://research.microsoft.com/Users/simonpj/papers/marktoberdorf/marktoberdo... my program uses both, together with ^Break handling, and ~10 C++ and Haskell threads running in parallel in my experience, exceptions are rarely required in Haskell program - i use them only to roll out when IO problems occur. OTOH, concurrency is very handy in Haskell/GHC - it's easy to create threads and communicate in reliable way, so it's a great tool to split algorithm into subtasks. and GHC lightweight threads make it very cheap - you may run thousands of threads. example program that uses one thread to produce numbers and another to print them is less than 10 lines long -- Best regards, Bulat mailto:Bulat.Ziganshin@gmail.com

Andrew Coppin

3:12 p.m.

Bulat Ziganshin wrote:

...

Hello Andrew,

definitive reading: "Tackling the awkward squad: monadic input/output, concurrency, exceptions, and foreign-language calls in Haskell" http://research.microsoft.com/Users/simonpj/papers/marktoberdorf/marktoberdo...

I've read it. Is everything described in that paper actually implemented now? (And implemented in exactly the same way as the paper says?)

...

in my experience, exceptions are rarely required in Haskell program - i use them only to roll out when IO problems occur.

Indeed. Somebody else mentioned Maybe; much cleaner, more intuitive solution.

...

OTOH, concurrency is very handy in Haskell/GHC - it's easy to create threads and communicate in reliable way, so it's a great tool to split algorithm into subtasks. and GHC lightweight threads make it very cheap - you may run thousands of threads. example program that uses one thread to produce numbers and another to print them is less than 10 lines long

It's nice that you can have millions of threads if you want to do something very "concurrent". What I tend to want is "parallel" - doing stuff that *could* be done in a single thread, but I want it to go faster using my big mighty multicore box. As I understand it, you have to do something "special" to make that happen...? While we're on the subject... has anybody ever looked at using muptiple processors on *networked* machines? Haskell's very pure semantics would seem quite well-suited to this...

Bulat Ziganshin

4:28 p.m.

New subject: Re[2]: Parallel + exceptions

Hello Andrew, Saturday, June 23, 2007, 7:12:52 PM, you wrote:

...

Is everything described in that paper actually implemented now? (And implemented in exactly the same way as the paper says?)

difference may be in subtle details. it just works for me :)

...

...
in my experience, exceptions are rarely required in Haskell program - i use them only to roll out when IO problems occur.

...

Indeed. Somebody else mentioned Maybe; much cleaner, more intuitive solution.

for me Maybe works great for exclusion of failed items (file not found, etc) while exceptions are used to transfer control over the many calling levels when we need to abort large piece of computation entirely

...

It's nice that you can have millions of threads if you want to do something very "concurrent". What I tend to want is "parallel" - doing stuff that *could* be done in a single thread, but I want it to go faster using my big mighty multicore box.

this isn't completely different things. threads' easy of use means that you can split your algorithm into parallel tasks here and there, improving CPU utilization

...

As I understand it, you have to do something "special" to make that happen...?

my own program spend most of its time in C threads, so i can't make exact answer to your query. afair, you need to use -threaded when compiling program, "+RTS -N2" for 2-core box when you run it - that's all. ask me if you need C threads too

...

While we're on the subject... has anybody ever looked at using muptiple processors on *networked* machines? Haskell's very pure semantics would seem quite well-suited to this...

i'm not sure, but this may be what you are looking for: http://f.loulergue.free.fr/ftp/bsmllib-0.25.tar.gz http://f.loulergue.free.fr/ftp/bsmllib-0.25_manual_html.tar.gz http://www.inf.ufes.br/~ffranzosi/BSPHlib-0.1.tar.gz http://f.loulergue.free.fr/ftp/minibsml-0.11.tar.gz -- Best regards, Bulat mailto:Bulat.Ziganshin@gmail.com

Felipe Almeida Lessa

11:23 p.m.

On 6/23/07, Andrew Coppin wrote:

...

It's nice that you can have millions of threads if you want to do something very "concurrent". What I tend to want is "parallel" - doing stuff that *could* be done in a single thread, but I want it to go faster using my big mighty multicore box. As I understand it, you have to do something "special" to make that happen...?

Have you seen this paper? http://www.macs.hw.ac.uk/~dsg/gph/papers/html/Strategies/strategies.html It's probably want you want, I think. -- Felipe.

Andrew Coppin

25 Jun 25 Jun

6:38 p.m.

Felipe Almeida Lessa wrote:

...

On 6/23/07, Andrew Coppin wrote:

...
It's nice that you can have millions of threads if you want to do something very "concurrent". What I tend to want is "parallel" - doing stuff that *could* be done in a single thread, but I want it to go faster using my big mighty multicore box. As I understand it, you have to do something "special" to make that happen...?

Have you seen this paper? http://www.macs.hw.ac.uk/~dsg/gph/papers/html/Strategies/strategies.html

It's probably want you want, I think.

Well, it certainly looks interesting... thanks.

Andrew Coppin

7:12 p.m.

Andrew Coppin wrote:

...

Felipe Almeida Lessa wrote:

...
Have you seen this paper? http://www.macs.hw.ac.uk/~dsg/gph/papers/html/Strategies/strategies.html

It's probably want you want, I think.

Well, it certainly looks interesting... thanks.

Does anybody know where I can find a version of this paper that actually has the diagrams in it?

dons＠cse.unsw.edu.au

24 Jun 24 Jun

7:57 a.m.

andrewcoppin:

...

Bulat Ziganshin wrote:

...
Hello Andrew,

definitive reading: "Tackling the awkward squad: monadic input/output, concurrency, exceptions, and foreign-language calls in Haskell" http://research.microsoft.com/Users/simonpj/papers/marktoberdorf/marktoberdo...

I've read it.

Is everything described in that paper actually implemented now? (And implemented in exactly the same way as the paper says?)

...
in my experience, exceptions are rarely required in Haskell program - i use them only to roll out when IO problems occur.

Indeed. Somebody else mentioned Maybe; much cleaner, more intuitive solution.

...
OTOH, concurrency is very handy in Haskell/GHC - it's easy to create threads and communicate in reliable way, so it's a great tool to split algorithm into subtasks. and GHC lightweight threads make it very cheap - you may run thousands of threads. example program that uses one thread to produce numbers and another to print them is less than 10 lines long

It's nice that you can have millions of threads if you want to do something very "concurrent". What I tend to want is "parallel" - doing stuff that *could* be done in a single thread, but I want it to go faster using my big mighty multicore box. As I understand it, you have to do something "special" to make that happen...?

While we're on the subject... has anybody ever looked at using muptiple processors on *networked* machines? Haskell's very pure semantics would seem quite well-suited to this...

MPI/cluster stuff has also been done a fair bit, but in these days SMP multicore machines are a hot topic. Start here: http://haskell.org/haskellwiki/Applications_and_libraries/Concurrency_and_pa... P.S. An awful lot of your questions are previously answered on the wiki :-) It's a good resource! -- Don

6587

Age (days ago)

6589

Last active (days ago)

List overview

Download

8 comments

5 participants

participants (5)

Andrew Coppin
Bulat Ziganshin
dons＠cse.unsw.edu.au
Felipe Almeida Lessa
Neil Mitchell