Re: [Boost-users] Several novice questions concerning threading

Subject: Re: [Boost-users] Several novice questions concerning threading
From: Ovanes Markarian (om_boost_at_[hidden])
Date: 2012-04-11 13:35:06

On Wed, Apr 11, 2012 at 6:39 PM, Master <master.huricane_at_[hidden]> wrote:

> Thank you very much :)
> for keeping track of my created threads i decided to use thread_groups and
> accumulate them there . i came up with sth like this :
>
>         boost::thread_group threadstore;
>
>         threadstore.add_thread(&thread);
>         threadstore.add_thread(&thread2);
>         BOOST_FOREACH(boost::thread t ,threadstore.threads)
>         {
>             cout<<t.get_id();
>         }
>
> As far as I can see the docs thread_group (
http://www.boost.org/doc/libs/1_49_0/doc/html/thread/thread_management.html#thread.thread_management.threadgroup)
does
not expose the public member threads. Even if it is there, it is an
implementation detail, which can change/(made private) with any release of
boost and your code will break. I suggest you to implement an approache
suggested by Vicente.

> well it didnt compiled ! before that , i tried using sth like this which
> failed nontheless .
>
>        vector<boost::thread> threadstore;
>
>         threadstore.push_back(thread);
>         threadstore.push_back(thread2);
>         BOOST_FOREACH(boost::thread t ,threadstore)
>         {
>             cout<<t.get_id();
>         }
>
> Thread group is just a management construct, it does not allow you to
iterate over the threads. BOOST_FOREACH supports STL like container.
thread_group does not provide begin/end iterators.

>
> im clueless of the cause .
> The threadstore has member called  :threads , which i dont know how to
> work around it , there is also another member named: m ,
>  which i have no clue of what it is !nor i know where it came form! or
> what i can do with it!
>
 i couldnt find any information on these members on the boost::thread
> documentation either
>
This an implementation detailed. It should be transparent for you, don't
use it.

> for that example i posted , i moved the mutex inside the loop and then
> used a sleep() method for couple of microseconds and got it working ( i
> mean now both threads seem to work as i expected them).
> but i want to know if we have sth like , lets say a timed_lock kind of
> lock !, so that a thread would only hold a mutex for a specified time , and
> when the timeslice presented in the timed_lock() passes , the
> affordmentioned thread releases the mutex
>
and thus other thread(s) can get that mutex and so there would be no need
> for a sleep() for a thread to wait till it time-slice finishes up and thus
> releases the mutex .the current timed_lock tries to obtain the mutex in a
> specified time , which is not the case .
>
I think you mixup smth. here. Generally speaking there are critical regions
in a parallel application. These regions must be protected by
synchronization objects to grant only a single thread modification for that
region. Now what you ask for: I know there is a long critical region, but
in the middle of the critical region I want a break and another thread
should run than. If it is so, make 2 regions, but there is no way to
implement smth like that in that simple manner. May be transactions, that
you interrupt the execution, roll back the calculated state, let others run
and start the calculation again.

> i remember i tried to use yield() for achieving such a possiblity
> (releasing mutex as soon as possible , i think it kinda worked ,i gave a
> similare result when i used sleep() )
>
How should it help, if the other thread wants in the exact critical section
which is locked? Yield just gives the remaining CPU time slice of the
thread to the scheduler. And scheduler might decide who runs next. It can
even happen, that this one thread runs again, if its prio the highest.

> here is the code which i wrote to actually speed up the sum action , which
> i think didnt give any speed ! would you see where the problem is ?
> //in the name of GOD
> //Seyyed Hossein Hasan Pour
> //Working with Boost::threads
> #define BOOST_THREAD_USE_LIB
> #include <iostream>
> #include <boost/date_time/posix_time/posix_time.hpp>
>
> #include <boost/thread.hpp>
> using namespace std;
>
> boost::uint64_t  i = 0;
> boost::uint64_t  sum=0;
> boost::mutex mutex;
>
> void IteratorFunc()
> {
>
>     for (i ; i<100000; i++)
>     {
>         mutex.lock();
>         sum+=i;
>        cout<<i<<"\t"<<boost::this_thread::get_id()<<endl;
>         mutex.unlock();
>         //boost::this_thread::sleep(boost::posix_time::microseconds(200));
>        boost::this_thread::yield();
>     }
>
> }
>
> int main()
> {
>
>     boost::posix_time::ptime start =
> boost::posix_time::microsec_clock::local_time();
>     boost::thread thread(IteratorFunc);
>     boost::thread thread2(IteratorFunc);
>
> //    boost::thread_group threadstore;
> //
> //    threadstore.add_thread(&thread);
> //    threadstore.add_thread(&thread2);
> //
> //    BOOST_FOREACH(boost::thread t ,threadstore.threads)
> //    {
> //        cout<<t.get_id();
> //    }
>
>     boost::posix_time::ptime end =
> boost::posix_time::microsec_clock::local_time();
>
>     thread.join();
>     thread2.join();
>
>     cout << "sum =\t" << sum<< "\t"<<end-start<<endl;
>     return 0;
> }
>
The speed up would rely on the lock-free implementation and dividing of
work into independent pieces. How sleep suppose to speedup smth. if in that
time period is nothing calculaded? CPU just stands still. Sum can be
speedup greatly. The keyword here is work stealing, first proposed in Cilk (
http://en.wikipedia.org/wiki/Cilk). Intel acquired Cilk. For C++ you can
use Intel Threading Building Blocks, which has a parallel_reduce algorithm.
Just download the tutorial:
http://threadingbuildingblocks.org/uploads/81/91/Latest%20Open%20Source%20Documentation/Tutorial.pdfand
take a look at chapter: 3.3.

>
> i also want to know if we have such a capability where i can specify the
> priority of a thread or a group of thread against the other threads .
>
You need to get the native handle and use the native OS functions.

> let me explain it little more, by that i mean , suppose we have couple of
> reader threads and one or two writer threads , is there any kind of
> possiblity that i can grant the writer thread(s) more priority in terms of
> accessing
> a resource (by obtaining the mutex more often ? - or the writer thread(s)
> deny access to readers in some cases ? ) ? or a part of memory ? ( e.g an
> array of some kind ? )
> if it is possible , how can i achieve such possibility ?
>
Boost provides basic reader/writer lock concepts. For more finer grained
concepts I think you will need implement this concept yourself. But usually
reader/writer is implemented the way that if a writer wants to enter the
critical section no other readers will enter it before the writer is done.
And writer has to wait until all currently active readers are done.
Actually the speedup here would be to increase the priority of readers to
the priority of the writer.

> is it possible that i can know if a thread was successful in doing what it
> was sent to ?
>
Actually, you should decouple threads from work and consider using futures
and thread pools. You are not interested in the thread's state, but the
result which was calculated in parallel. So you wait until the result is
available and verify it.

> can i specify that a thread or a group of thread execute in a specific
> order ? for example thread one must always execute first and thread two
> must always follow thread one . do we have such a thing ?
>
Again, future pattern would help. You put work items in the queue and they
are calculated in parallel but taken from the queue in a predefined order.
You can also use a priority queue to sort the items according their
priority when those are queued. On the other hand: You ask here for
sequential execution. Why do you need threads than?

> do i have any means of talking to threads ? checking the satus of a
> specific thread ? or group of threads ?
>
What status would you like to check?

> can i know by any means , that which threads are blocked and which are not
> ?
>
Blocked in terms of what? Waiting to enter the critical section or are
currently not running?

> Thank you so much for your time and please excuse me for such newbish and
> yet long questions .
> i really do appreciate your help and time :)
> Regards
> Hossein
>
>
>
> On Wed, Apr 11, 2012 at 7:38 PM, Vicente J. Botet Escriba <
> vicente.botet_at_[hidden]> wrote:
>
>>  Le 11/04/12 13:31, Master a écrit :
>>
>> Hello all .
>> i am a newbie to the boost community . i recently started learning about
>> threads in boost . now there are some questions i would like to ask :
>>
>> Welcome.
>>
>>  1.where can i find examples showing practical uses of boost::thread
>> features?
>>
>> The documentation doesn't contains too much examples. You can take a look
>> at the libs/thread/example and tutorial directories :(
>>
>>  2.how can i get all threads ID issued by me in my app?
>>
>> No direct way other that storing them in a container. What is your use
>> case?
>>
>>  3.how can i iterate through running threads in my app ?
>>
>> No direct way other than storing a thread pointer in a container. What is
>> your use case?
>>
>>  4.is there any kind of means to get all the running threads using boost
>> library? if it does whats the calss? if it doesnt how can i do that?
>>
>> See above. I think that you need to specialize the thread class so that
>> it inserts a handle to the created thread on a container at construction
>> time and remove it at destruction time.
>>
>>  5.can i resume a thread after pausing it ? ( how can i pause a thread? )
>>
>> Boost.Thread doesn't provide fibers or resumable threads. There is
>> Boost.Fiber for that purpose (not yet in Boost).
>>
>>  6. how can i share a variable between two or more threads , suppose i
>> have a loop , i want two threads to simultaneously iterate through it , if
>> thread1 counted to 3, thread2 continues it from 4 and so on . ?
>> i already tried
>>
>> You need to protect the access to the loop index variable 'i' with a
>> mutex as you did with sum.
>>
>> HTH,
>> Vicente
>>
>>  ------
>>
>>> what is wrong with my sample app ?
>>> #include <iostream>
>>> #include <boost/thread.hpp>
>>> using namespace std;
>>> using namespace boost;
>>>
>>> mutex bmutex;
>>> int i=0;
>>> int sum=0;
>>> void IteratorFunc(int threadid)
>>> {
>>>              for (  ; i<25 ; i++)
>>>             {
>>>                     lock_guard<mutex> locker(bmutex);
>>>
>>> cout<<"\t"<<threadid<<"\t"<<this_thread::get_id()<<"\t"<<i<<"\n";
>>>                       sum+=i;
>>>             }
>>> }
>>>
>>> int main()
>>> {
>>>     //boost::posix_time::ptime start =
>>> boost::posix_time::microsec_clock::local_time();
>>>
>>>     thread thrd(IteratorFunc,1);
>>>     thread thrd2(IteratorFunc,2);
>>>
>>>     cout<<sum;
>>>     thrd.join();
>>>     thrd2.join();
>>> }
>>
>>