Systems Performance

About the book

Brendan Gregg's handbook covers performance analysis for operating systems and the software running above them, with attention to enterprise deployments and cloud environments. It treats performance as an empirical discipline: form a hypothesis, gather data, and let measurements decide. The scope runs across the operating system, from CPU scheduling and memory management to file systems, disks, and networks, along with the tools used to observe those layers.

Gregg's background in tracing and profiling tools shapes the method. The book builds a vocabulary for reasoning about latency, throughput, utilization, and saturation, then shows how to move from a reported problem to the subsystem responsible. The cloud material matters because shared and virtualized infrastructure changes what a measurement means and who controls the underlying hardware.

Methodology comes first and components second. That order reflects a view that tool fluency without a diagnostic method produces guesswork, while a sound method can be applied to systems the reader has never seen.

How to read it

Read the methodology material closely before the component chapters. The early sections establish how to interpret counters and traces, and those habits carry through the rest of the book. Skim the tool references on a first pass and return to them when a specific problem demands them.

Expect dense technical prose and many diagrams, tables, and command examples. Once the method is clear, the component chapters can be read out of order. Keep a test system nearby if possible. The material rewards hands-on checking more than passive reading, and some tools differ across operating system versions.

Who it suits

This suits system administrators, site reliability engineers, and developers who troubleshoot production software and want a structured way to locate bottlenecks. It also helps architects who need to reason about capacity and latency in cloud environments. Readers with some operating systems background will get the most from it, since the component chapters assume familiarity with processes, memory, and I/O.

Reading considerations

The 2013 edition predates later changes in Linux tooling and cloud platforms, so specific commands and defaults may have moved on. Treat tool details as a starting point and verify against current documentation. The book is long and reference-like. Readers wanting a short primer should look elsewhere first.