Re: [Boost-users] [Threads] How to keep number of running thread constant, performing different operations

17 Jun 2011

      On Fri, Jun 17, 2011 at 12:20:31PM +0200, Ovanes Markarian wrote:
...
On Fri, Jun 17, 2011 at 11:10 AM, Lars Viklund  wrote:
...
On Fri, Jun 17, 2011 at 10:12:47AM +0200, Ovanes Markarian wrote:
...
On Fri, Jun 17, 2011 at 9:05 AM, Alessandro Candini 
wrote:
[...]
For managing the N threads, you can use a thread_group.
Yes, but than you must implement and debug your own thread pool pattern.
Thread group only creates a number of threads, but you need to implement the
producer/consumer problem to retrieve work tasks from each thread from the
queue. Now there is a choise to either use a ready thread pool or implement
it yourself... AFAIK Boost has no official  thread pool
implementation, which is not in internal impl namespace and guarantees the
interface.
We have thread_group, we have Asio. One needs nothing else. I think
you're misunderstanding what I suggested.

I do not consider doing this remotely bothersome, particularly as you
only need features present in Boost itself.

It should need no debugging, as it's practically impossible to get
wrong. If you for some reason need completion notifications from your
workers, have that as part of your task, or wrap your functor in
something.

---8<---
asio::io_service io;
scoped_ptrasio::io_service::work work(new asio::io_service::work(io));

boost::thread_group tg;
auto N = boost::thread::hardware_concurrency();

for (unsigned i = 0; i < N; ++i)
  tg.create_thread(boost::bind(&asio::io_service::run, boost::ref(io));

io.post(some_task);
io.post(some_other_task);

// invoke io.post whenever you want to enqueue something.

// time passes, shutdown time has arrived

work.reset();
tg.join_all();
---8<---

-- 
Lars Viklund | zao@acc.umu.se