In the first part of this blog, we described how we offloaded almost all I/O operations to I/O threads, thereby freeing more CPU cycles in the main thread to execute commands. When we profiled the execution of the main thread, we found that a considerable amount of time was spent waiting for external memory. This was not entirely surprising, as when accessing random keys, the probability of finding the key in one of the processor caches is relatively low. Considering that external memory acc...
