Conflicts in Cache Behavior Prediction

Memory transfers are the performance bottleneck of many applications due to poor data locality and limited memory bandwidth. Code refactoring for better data locality can improve cache behavior, leading to significant performance boosts. Reuse distance, a measure of data locality, is useful in identification and optimization of hot code regions exhibiting poor data locality.

Benefits

Successful completion of the bachelor's thesis enables you to enter responsible positions, for example in the areas of artificial intelligence, gaming and high-frequency trading. The ability to systematically identify and resolve performance bottlenecks is a highly sought-after skill.

Performance matters! During this thesis you can:

• Gain valuable hands-on experience in high-performance computing and computer architecture
• Have the opportunity to work with state-of-the-art hardware architectures
• Deepen your knowledge in the C/C++ programming languages
• Get in touch with the Future Computing Group at the Leibniz Supercomputing Centre (LRZ)

Motivation and Background

Reuse distance is defined as the number of unique memory locations referenced between a pair of references to the same memory location. On the granularity of cache lines, reuse distance can model spatial and temporal locality to assess cache behavior of applications. Assuming a fully associative cache with least recently used (LRU) replacement policy, predicting cache behavior with reuse distance is exact.

However, several cache-specific details are ignored by reuse distance:

• Limited cache associativity (conflict misses)
• Deviations from LRU replacement policy
• Hardware prefetching

On top, even another level of complexity is introduced by cache-sharing in today's multi-core systems.

Goals and Tasks

In this thesis, you will explore the accuracy of cache behavior prediction with reuse distance in irregular applications. As an example application, you will use sequential sparse matrix-vector multiplications (SpMV), a ubiquitous kernel, for instance in simulations and graph algorithms.

In the context of this thesis, you will:

• Adjust an existing C++ reuse distance algorithm to include cache associativity
• Implement a simple pseudo-LRU (bit- / tree-based) cache simulator
• Design and conduct experiments with sequential (optional: parallel) SpMV on several state-of-the-art hardware architectures of the BEAST system
• Compare four cache-behavior prediction approaches with cache-related hardware performance events obtained from SpMV execution:

1. Reuse distance
2. Associativity-aware reuse distance
3. pseudo-LRU
4. Cachegrind (a Valgrind tool)

Prerequisites

• Basic understanding and interest in computer architecture and multi-core memory hierarchies
• Basic knowledge in high-performance and parallel computing (Parallel and High-Performance Computing or equivalent)
• Proficiency in Linux, and C/C++ (Systempraktikum or equivalent)

Starting Literature

• Beyls and. D’Hollander Reuse Distance as a Metric for Cache Behavior
• Hennessy and Patterson Computer Architecture a Quantitative Approach (5th ed.)

Organisatorisches

• Aufgabensteller: Prof. Dr. D. Kranzlmüller
• Dauer der Bacheorarbeit: 3 Monate
• Anzahl Bearbeiter: 1
• Betreuer: Sergej Breiter
• Sprache: Deutsch oder Englisch