线程池和数据关联

Question

I am designing an OMS and trying to use multi-threading techniques to speed up processing.

An easy way is to allocate threads statically: one thread for all securities starting with a letter. Obviously some of them could be very busy while others have nothing to do. And we don't know in advance what stocks users will enter.

So I turn to thread pool. Now time is likely to be distributed more evenly among threads. But there is an overhead to be dealt with: I have to lock each order, which certainly is not pleasant.

Any idea about how to minimize lock time and achieve higher throughput?

Thanks a lot for suggestions.

PS, it'a Unix environment. Language used is C++

Answer 1

Generally, thread per task is very bad decision. The best way is separating tasks into logical groups with own thread pools (or, may be with 1 thread per task). Also you should organize communication between tasks for example by queries. Do you looking throw existed thread pool implementations like threadpool ?