Memory Optimization Techniques

Design and implement a high-throughput, low-latency C++ service that receives a real-time stream of market-data ticks (symbol, price, size, timestamp) and must continuously publish the Volume-Weighted Average Price (VWAP) for every symbol over the last N minutes (sliding window). Your implementation must minimize both CPU and memory footprint while handling 5M+ ticks/second on a single NUMA node. You may assume that the number of distinct symbols is bounded by 50,000 at any moment and that the window length N is provided once at start-up and never changes. You are required to justify every data-structure and allocation choice with cache-performance numbers and to show how your design behaves as the working set exceeds L3 cache size.