“Memory Pools and Region-based Memory Management
Memory pools and region-based memory management allow you to improve your program performance by avoiding unnecessary memory freeing calls. Moreover, pure memory pools gain even more performance due to simpler internal mechanisms. The techniques are widely used in Web servers and using them you can do following (pseudo code for some imaginary Web server):
http_req->pool = pool_create();
while (read_request & parse_request) {
http_req->hdr[i] = pool_alloc(http_req->pool);
// Do other stuff, don’t free allocated memory.
}
// Send the request.
// Now destroy all allocated items at once.
pool_destroy(http_req->pool);
This reduces number of memory allocator calls, simplifies its mechanisms (e.g. since you don’t free allocated memory chunks it doesn’t need to care about memory fragmentation and many other problems which common memory allocators must solve at relatively high cost) and makes the program run faster.
Probably, you’ve noticed that we call pool_alloc() in the example above without specifying allocation size. And here we go to the difference between memory pools and region-based memory management: memory pools can allocate memory chunks with only one fixed size while region-based memory allocators are able to allocate chunks with different size. Meantime, both of them allow you not to care about freeing memory and just drop all the allocated memory chunks at once…”
http://natsys-lab.blogspot.com.br/2015/09/fast-memory-pool-allocators-boost-nginx.html
