LC

Fix False Sharing Bug: Two Threads Updating Adjacent Array Elements

medium2/8

Demonstrate Row-Major vs Column-Major Matrix Traversal Structure of Arrays vs Array of Structures Benchmark

Language

Compiler

2.0s Limit 256MB

Implementation

Demonstrate and fix false sharing in multithreaded code.

Requirements:

Create two threads incrementing separate counters
Show false sharing without padding (slow)
Fix with cache line padding (fast)
Measure performance difference (10-100x speedup)
Use pthreads for thread creation
Print performance comparison
Show cache line size (typically 64 bytes)
Use attribute((aligned(64))) or manual padding
Include proper thread synchronization
Demonstrate understanding of cache coherence

Why This Matters

False sharing has caused real, hard-to-spot production slowdowns — the Java Disruptor library (LMAX) was built explicitly around padding ring-buffer sequence counters to their own cache line after engineers traced a throughput cliff to exactly this bug. Go's runtime and many lock-free queue implementations still pad hot counters to 64 or 128 bytes for the same reason, since the fix is nearly free but the bug is brutal to diagnose without a profiler.

Hints

0 / 3 revealed

Validation Suite

main.c

Workspace 1.0