Efficient Characterization of Hidden Processor Memory Hierarchies

摘要

A processor's memory hierarchy has a major impact on the performance of running code. However, computing platforms, where the actual hardware characteristics are hidden from both the end user and the tools that mediate execution, such as a compiler, a JIT and a runtime system, are used more and more, for example, performing large scale computation in cloud and cluster. Even worse, in such environments, a single computation may use a collection of processors with dissimilar characteristics. Ignorance of the performance-critical parameters of the underlying system makes it difficult to improve performance by optimizing the code or adjusting runtime-system behaviors; it also makes application performance harder to understand. To address this problem, we have developed a suite of portable tools that can efficiently derive many of the parameters of processor memory hierarchies, such as levels, effective capacity and latency of caches and TLBs, in a matter of seconds. The tools use a series of carefully considered experiments to produce and analyze cache response curves automatically. The tools are inexpensive enough to be used in a variety of contexts that may include install time, compile time or runtime adaption, or performance understanding tools.