San Francisco: Managing Memory Maliciously

The majority of CPU cycles in real world systems are spent idly waiting on memory operations. Optimizing memory performance is a tricky art because of the large number of hardware and software components involved: virtual addressing, caching, out-of-order execution, allocation strategies and object lifetime management, to name a few.

My goal is simple, but challenging: to show you how memory works and to teach you the mindset necessary for tackling memory-related performance bottlenecks. To that end, I'll dig into my bag of tricks and share some of the more interesting techniques I've used over the years to optimize C++ software with memory troubles.

Bryce Adelstein Lelbach is a researcher at Lawrence Berkeley National Laboratory (LBNL) working on new parallel programming models for exascale and post-Moore architectures. Bryce is a developer of the HPX C++ runtime system; a Boost contributor; an organizer for the C++Now and CppCon conferences; and LBNL's representative to the C++ standards committee.

Level of Expertise: Intermediate to Advanced