Finding performance bottlenecks with Pyroscope and Alloy: An example using TON blockchain

This post explores how continuous profiling with Grafana Alloy and Pyroscope was used to analyze and visualize performance bottlenecks in TON blockchain C++ optimization contest submissions.

• •Alloy's pyroscope.ebpf component uses eBPF-based CPU profiling to capture system-wide data without modifying application code
• •Replacing the default OpenSSL SHA256 with the SerenityOS implementation yielded ~2% speedup; consolidating SHA256 feed calls gave 3.5% overall and 20% improvement in DataCell::create
• •Swapping std::map for std::unordered_set in CellStorageStat::add_used_storage reduced O(log n) tree lookups to O(1) hash lookups, resulting in ~10% speedup
• •Ed25519 signature verification was optimized with x86_64 handwritten assembly to achieve ~1.5% speedup
• •