Question

I am researching for an architecture for creating an application in C++ to process multiple inputs at the same time in different cores. Each input is processed in a single core simultaneously. Each of the process on a core, the same filters will be processed through. For example: filter1.apply(), filter2.apply() and filter3.apply(). The processes are illustrated for 4 cores for 4 inputs as below:

[core 1]   [core 2]    [core 3]   [core 4]
   |          |            |         |     
   V          V            V         V
  input1     input2      input3     input4
   |          |            |         |    
   V          V            V         V 
 filter1    filter1     filter1    filter1
   |          |            |         |    
   V          V            V         V 
 filter2    filter2     filter2    filter2
   |          |            |         |    
   V          V            V         V 
 filter3    filter3     filter3    filter3
   |          |            |         |    
   V          V            V         V 
 output1   output2      output3    output4

I dont know which architecture or design pattern is suitable for this. It would be great if you give me some artifacts (documents or sample application) to read further.

Thanks in advance.

Answer 1

Usually thread pools are used to implement such designs. They scale virtually infinitely for independent processes. You can find easy implementations in TBB and PPL. They offer concurrency-related versions of many Standard algorithms and containers. For example, in this sample I’ve used concurrent_vector, which is a vector-like container which can be safely mutated from several threads concurrently, and parallel_for_each, which runs the function on many threads at once.

concurrent_vector<output> outputs;
std::vector<input> inputs;
parallel_for_each(inputs.begin(), inputs.end(), [](input& input) {
    outputs.push_back(filter3(filter2(filter1(input))));
});