LC

Demonstrate False Sharing and Cache Coherency Overhead

medium5/9

Implement LRU Eviction Policy for Cache Replacement Implement Cache-Oblivious Matrix Multiplication

Language

Compiler

2.0s Limit 256MB

Implementation

Demonstrate false sharing and cache coherency overhead.

Requirements:

Create two-thread benchmark
Measure throughput with false sharing
Fix with padding/alignment
Show performance difference
Explain cache coherency protocol
Demonstrate MESI states
Use perf counters to show misses

Why This Matters

False sharing is a real, silently catastrophic bug: Java added the @Contended annotation and the LMAX Disruptor pattern pads its ring-buffer cursors specifically to prevent it, and the Linux kernel's per-CPU variables are deliberately cache-line-aligned for the same reason. Any multi-threaded counter or lock-free data structure that ignores this, from database engines to game engines, can silently lose 10-20x throughput exactly as you measure here.

Hints

0 / 3 revealed

Validation Suite

main.c

Workspace 1.0