N4468: On Quantifying Memory-Allocator Strategies Page 1 of 34 Doc No: N4468 Date: 2015-04-11

Performance requirements drive many of our difficult design choices. Memory management is an area where such choices can have surprising and far-reaching effects. Although performance of global allocators has improved markedly in recent years, use of local memory allocators can still provide substantial runtime (and other) benefits. The key to the effective use of memory allocators is knowing if and when to use which allocator and why. To be able to make reasoned recommendations regarding the use of local memory allocators, we must first understand where and how they can affect runtime performance. We have identified several ways to characterize how systems can challenge a global allocator, and how they may benefit by applying a well-chosen local allocator. In order to develop optimal criteria for how to choose where and how to apply a local allocator, we need to obtain objective measurements. We have identified several usage patterns, which we have encoded into benchmarks to identify precisely where local allocators do (and do not) provide substantial benefits. This paper presents our preliminary “raw ” initial quantitative results (with relatively little analysis) in the hope of stimulating informed discussion. Implementations of standard allocators (and others) are freely available today – along with usage examples – in Bloomberg’s open-source distribution of the BDE library a