One Billion Row Challenge

One Billion Row Challenge. The problem easily fits in your head, allowing you to think about an initial algorithm. To discover which parts of the code are slow two methods are used : flamegraph and perf. The analysis demonstrates several well-known optimization techniques, such as:

• Using bytes instead of UFT8 string
• Implementing faster hash functions, by dropping some characteristics (for example, resistance against HashDoS)
• Utilizing Memory Map
• Leveraging SIMD
• Adding parallelization.

While solving performance problems often involves trial and error, it’s refreshing to read about performance optimization in a language other than C or C++.

Fixing undefined behaviour: back to 1.56s

— https://curiouscoding.nl/posts/1brc/#fixing-undefined-behaviour-back-to-1-dot-56s

Bonus: The command lstopo will output an image representing your system’s topology.

Notes (for an Intel i7-12700T):

• L1 cache is split into d (for data) and i (for instructions).
• Intel E-cores (Efficiency) CPUs share L2 cache, and L1 cache split is different than P-cores (Performance)