Networking · Cluster
Network Performance
Latency budgets, buffering, and measurement.
- 01 Decomposing Path Latency into Propagation, Serialization, and Queueing The four terms that sum to the delay across a network path, how each scales with distance, link rate, and load, and which ones a sender can actually influence.
- 02 Why Queueing Delay Rises Sharply Near Link Utilization How the shape of the M/M/1 waiting-time curve explains the knee in latency-versus-load graphs, and why utilisation targets are set where they are.
- 03 Standing Queues and the Buffer Sizing Trade-Off Why deep buffers turn into permanent added latency, how the bufferbloat diagnosis differs from ordinary congestion, and what the bandwidth-delay product says about how much buffer is needed.
- 04 Serialization Delay and the Cost of Small Packets Why shrinking packets reduces queueing delay but raises packet rate, header overhead, and per-packet forwarding cost, and where the crossover sits.
- 05 Counting Round Trips in Connection Setup How handshakes, name resolution, and security negotiation add serial round trips before the first byte of payload, and which of them can be overlapped.
- 06 What ICMP Echo and TTL Probing Actually Measure The difference between the forward and return paths a traceroute samples, why hop times are often wrong, and what rate limiting on the control plane does to the numbers.
- 07 One-Way Delay and the Clock Offset Problem Why a one-way delay measurement needs synchronised clocks, how a constant offset biases the result, and how two-way exchanges bound the error.
- 08 Active Queue Management and Latency Under Load How tail-drop buffers convert delay into loss, what controlled-delay algorithms change about the signal a sender receives, and why their tuning constants are what they are.
- 09 Priority and Fair Queueing for Mixed Traffic How strict priority starves low classes, how deficit round robin allocates bandwidth by packet size, and how the two combine on an access link.
- 10 Measuring Latency Under Load Without Fooling Yourself How to design a delay measurement that separates fixed path delay from queueing, which load patterns expose bufferbloat, and why idle-path numbers mislead.