Performance Engineering

Build real systems software with Odin. This practical guide takes you from the language basics to memory, concurrency, networking and performance. Learn how Odin works under the hood while building parsers, allocators, simulations and more, with complete examples and clear comparisons to C, C++, Rust and Go.

Your C++ gateway is fast. It still loses races. The cost sits below your code, in kernel network stacks, PCIe round trips, and OS scheduling jitter that no profiler shows and no rewrite removes.This book moves market data ingestion and a bounded slice of decision logic into FPGA fabric, cutting a typical 3 to 7 microsecond software path to roughly 0.7 to 1.1 microseconds, with every number measured end to end, not assumed.

What really makes software fast? This book takes you inside the modern CPU to show what happens to your code at the hardware level. Learn how to measure performance, find bottlenecks and turn slow code into faster code using practical techniques, real tools and real hardware.

Master the ideas behind Java concurrency, from threads, race conditions, and synchronization to CompletableFuture, virtual threads, and structured concurrency. With practical patterns, runnable examples, and a companion GitHub repository, this book helps you move from understanding concurrency to applying it in real programs.

Your Greeks change when you move from 32 cores to 64. Your prices drift after a compiler upgrade. Neither is noise. Both are bugs, and this book shows how to build C++ pricing engines that cannot have them.

What really happens inside a database when queries run, data is stored and performance starts to suffer? This book takes you behind the scenes of SQLite, PostgreSQL and modern storage engines, showing how they work through real code, practical examples and performance analysis. Learn to find bottlenecks, make better design decisions and build faster, more reliable applications.

Modern computer systems are complex, and getting the best performance out of them takes more than writing efficient code. This book takes you inside the hardware, from processor pipelines and cache hierarchies to parallel computing and acceleration. Learn how to measure, understand and eliminate performance bottlenecks with practical techniques grounded in real-world engineering.

What if you could make massive Mixture-of-Experts models smaller without retraining them from scratch? This book takes you from the fundamentals of expert pruning to practical implementation, covering proven methods, working code and real-world deployment challenges. Learn how to identify redundant experts, preserve model performance and build pruning solutions that work in production.

Python was never built to be the fastest language, but the world it now powers demands more speed than ever. As JIT compilation pushes Python into new territory, old trade-offs are coming back into focus. What does it take to make a language faster without losing what made it popular in the first place?

Across every chapter in this book, we'll look at real software, break it down and put it back together again so we can use it. We'll encounter a stripped binary, a scrambled algorithm, an undocumented file format, a mobile app protecting its backend, a web client, a hardened sample that fights back, and a live protocol whose secrets only exist while it's running. We're not stopping at one clever trick either. We'll get the job done to a standard that's verifiable and someone else can trust.

Your matching engine passed every test. Your risk engine rejects every bad order. Your FPGA gateway hits its latency target on the bench.On day one, all four components ran together against live order flow, and the system breached its exposure limit by several multiples.No component was wrong. The seams between them were.This book is about those seams.

Your CPU has been waiting.Every numeric loop you have ever shipped processed one value at a time on hardware that was built to process sixteen simultaneously. The gap between what your code does and what the silicon is capable of is not a compiler limitation or a language restriction. It is a knowledge gap, and it has a precise cost you can measure in cycles per element.This book closes it.You will learn to read hardware the way the processor reads it: in lanes, in registers, in parallel. You will understand why your data layout is the first performance decision and the last one most engineers make. You will write branchless conditional logic that executes at full vector throughput, navigate AVX-512’s predication model without triggering the frequency trap that makes wider vectors slower, and build ARM NEON kernels that exploit interleaved memory instructions x86 has no equivalent for.By the time you reach the final chapter, you will have the complete toolkit: a portable abstraction layer that compiles to native instructions with zero overhead, a compiler collaboration discipline that turns the auto-vectorizer from an unreliable guess into a predictable collaborator, a benchmarking methodology that measures real throughput instead of measuring the speed of nothing, and a fat binary deployment architecture that ships the right implementation to every machine in your fleet without a separate binary per hardware class.The hardware is fast. It has always been fast. This book teaches you to use it.

We will define the developer contract first, and we will then build the platform underneath it together, step by step, using decisions that hold up in regulated environments and in small startups alike. Along the way, we will provision clusters, wire delivery, publish a developer API, enforce policy and ship observability, and we will do all of it where you can see every command, every manifest and every failure. Nothing here stays theoretical, because platform problems only show their true shape where an abstraction meets a real cluster.

Build a real LLM inference engine from the ground up in portable C. Learn how transformers work, load SafeTensors weights, implement KV caching and optimize inference on the CPU. By the end, you will run TinyLlama locally and generate text with a complete working runtime.

What does it take to run an LLM on hardware that barely has enough memory to breathe? This practical guide takes you from transformer math and model compression to inference runtimes, benchmarking and real-world deployment, with hands-on case studies for building fully offline LLM systems on tiny devices.