Thread-Per-Core Buffer Management for a modern Kafka-API storage system Superscalar CPUs with wide GB/s memory, NVMe access times in the order of 10-100’s of microseconds, necessitates a new buffer management for low latency storage systems. By Alexander Gallego on September 29, 2020 11 min read As I have previously observed, software does not run on category theory, it runs on superscalar CPUs with wide, multi-channel GB/s memory units and NVMe SSD access times in the order of 10-100’s of microseconds. The reason some software written a decade ago - on a different hardware platform - feels slow is because it fails to exploit the advances in modern hardware.