我应该使用一个线程来更新数组中每个对象的值吗

Question

I am doing mutual information for 50 words for a big corpus. I need to count the concurrence for those target words and the neighbour words for each of them.

There is an object array with size of 50, one for each word.

I have a thread to read lines. When there is a new line comes in, I need to update the values of all those objects.

I want to write in a multithread way.

Should I use 50 threads, and each thread is responsible for one object in this array?

If I use one thread for each object, should I use 50 queues to store the input lines, and one queue for each thread?

I am thinking Producer-Consumer pattern, but it looks a little different where there is just one queue...

Answer 1

Multithreading is most beneficial when running on multi-core/multi-cpu machines. The number of Threads should be inferred from the number of available cores on the machine. If 5 cores are available, 5 Threads should be spun off. Any more threads and you could start loosing efficiency due to CPU context switch.

Here's an example of how you could do that:

int coreCount = Runtime.getRuntime().availableProcessors();
for(int i = 0; i < coreCount; i++) {
  Thread thread = new YourThreadClass();
  thread.start();
}

Answer 2

It is a bit confusing what your intention is. I'll try to discuss it in general. It is not advisable to have more threads running than you have CPU cores (some argue it is twice that number), because after that, you don't gain any more execution speed, but indeed pay a penalty. You should also consider what the effect of your "new line" input is. If all objects are updated in the same manner, there is no need for that many queues, one should be sufficient. If those objects are updated separately, at non-depending times, you could implement it as you suggested 50 threads with 50 queues, but if the update is uniform, you should consider finding a suitable solution with less threads (perhaps there is some logical connection between your objects, which could help you group some of them in fewer threads)