APEX‐Map: a parameterized scalable memory access probe for high‐performance computing systems

Abstract

SUMMARYThe memory wall between the peak performance of microprocessors and their memory performance has become the prominent performance bottleneck for many scientific application codes. New benchmarks measuring data access speeds locally and globally in a variety of different ways are needed to explore the ever increasing diversity of architectures for high-performance computing. In this paper, we introduce a novel benchmark, APEX-Map, which focuses on global data movement and measures how fast global data can be fed into computational units. APEX-Map is a parameterized, synthetic performance probe and integrates concepts for temporal and spatial locality into its design. Our first parallel implementation in MPI and various results obtained with it are discussed in detail. By measuring the APEX-Map performance with parameter sweeps for a whole range of temporal and spatial localities performance surfaces can be generated. These surfaces are ideally suited to study the characteristics of the computational platforms and are useful for performance comparison. Results on a global-memory vector platform and distributed-memory superscalar platforms clearly reflect the design differences between these different architectures.

How to fit program footprint curves

“…Denning surveyed analytical models of locality [1]. Recently, Ibrahim and Strohmaier uses an analytical model in APEX-Map for analyzing the miss rate curve [3,5]. Ding and Zhong [2] used linear fitting to find the correlation between the input size and the reuse distance.…”

FractalMRC: Online Cache Miss Rate Curve Prediction on Commodity Systems

“…A number of approaches have been proposed to generating shared cache MRCs [15,[28][29][30][31][32][33]. The common technique using these approaches is to calculate the LRU stack distance (also called reuse distance) which is first proposed by Mattson et al [15].…”

“…A variety of schemes or models thus have been created to reduce this overhead. Ibrahim et al presented a mathematical equation to describe the relation between their Apex-Map framework [32] and reuse distance distributions [33]. To compute a reuse distance distribution, this approach needs to use Valgrind to collect the memory trace of the code of interest, which still incurs high runtime overhead and storage cost.…”

Software probes: towards a quick method for machine characterization and application performance prediction

“…Therefore, the initial step of our studies is the analysis of memory access patterns of application's phases, and the subsequent creation of a synthetic tunable probe that could mimic this memory pattern. Both [18] and [20] are different approaches of the development of a synthetic probe that can represent memory access. We aim to create a probe who is able to represent not only the memory access precisely, but as well to mimic another factors important to represent the performance of an application.…”