Network Performance Metrics
Network performance is characterized by a small set of metrics — throughput, delay, jitter, and loss — that together determine how fast and how reliably data moves between hosts and how well a network serves its applications.
Definition
Network performance metrics are the quantitative measures — principally throughput, delay (latency), delay variation (jitter), and packet loss — used to characterize how a network or path delivers data.
Scope
This topic covers the quantitative measures of network performance. It treats the four components of packet delay (processing, queueing, transmission, and propagation), end-to-end and round-trip latency, throughput and its relationship to bottleneck links, jitter (delay variation), and packet loss. It also covers the bandwidth-delay product and how these metrics shape application experience. It excludes the protocol mechanisms, such as congestion control, that respond to these conditions, which are treated in the transport-layer area.
Core questions
- What are the four components of packet delay, and which dominate in different settings?
- How does throughput relate to the bottleneck link along a path?
- What is the difference between bandwidth, throughput, and goodput?
- How do jitter and packet loss arise, and which applications are sensitive to them?
- What is the bandwidth-delay product, and why does it matter?
Key concepts
- throughput and goodput
- bandwidth
- latency (one-way and round-trip)
- processing, queueing, transmission, propagation delay
- jitter (delay variation)
- packet loss
- bottleneck link
- bandwidth-delay product
Key theories
- Nodal delay decomposition
- The delay a packet experiences at each node is the sum of processing, queueing, transmission, and propagation delays; queueing delay is the variable, load-dependent component, while propagation depends on distance and transmission on link rate and packet size.
- Throughput and the bottleneck link
- End-to-end throughput is limited by the lowest-capacity (bottleneck) link along a path and by contention from competing traffic, so adding capacity elsewhere does not help unless it relieves the bottleneck.
- Bandwidth-delay product
- The product of a link's bandwidth and the round-trip delay measures how much data can be 'in flight' on a path; it sets the window size needed to keep a high-speed, long-distance link fully utilized.
Clinical relevance
Performance metrics directly translate into user experience and engineering decisions: low latency is critical for interactive applications such as gaming and video calls, high throughput matters for bulk transfers and streaming, and low jitter and loss are essential for real-time media. Measuring and budgeting these metrics underlies capacity planning, service-level agreements, and the diagnosis of slow or unreliable connections.
History
Queueing theory, developed by Leonard Kleinrock and others, gave the early analytical foundation for reasoning about delay in packet networks. As the Internet grew and diversified into latency-sensitive and bandwidth-hungry applications, the systematic treatment of throughput, delay, jitter, and loss became standard in networking texts and central to measurement tools and service guarantees.
Key figures
- Leonard Kleinrock
- James F. Kurose
- Larry L. Peterson
Related topics
Seminal works
- kurose2021
- peterson2021
Frequently asked questions
- What is the difference between bandwidth and throughput?
- Bandwidth is the maximum data rate a link or path can theoretically carry, while throughput is the rate actually achieved by a given transfer, which is limited by the bottleneck link, protocol overhead, and competing traffic. Goodput further excludes retransmissions and protocol headers to count only useful application data.
- Why does my connection feel slow even with high bandwidth?
- Perceived slowness often comes from latency, jitter, or loss rather than bandwidth. Interactive tasks depend on round-trip delay, so a high-bandwidth but high-latency or lossy path can feel sluggish, especially for real-time applications, even though large downloads might still complete quickly.