Introduction to network metrics

What are metrics?

Metrics are a quantitative and qualitative way to verify a desired behavior.  For instance many operators like to measure their "uptime", a count of how often services are available for users.  Network metrics are similar, and are related to desired outcomes.  Some measurements include:

  1. Network capacity can be described as "how much" traffic can cross a given link, segment, or aggregated path
  2. Network utilization is a measurement of how much of the capacity is currently in use
  3. Throughput (sometimes called achievable bandwidth) is a measure of how much of the network can be used at a given time.  This can be thought of as a form of inverse of the utilization, and is often measured for paths versus individual segments
  4. Round trip latency is the measure of how long data takes to travel between two hosts
  5. One way latency is a measure of a single direction between hosts
  6. Packet loss is a measurement of how many packets are dropped (reason agnostic) on a network segment or path
  7. Packet duplication is a measure of how many packets are duplicated (reason agnostic) on a network segment or path
  8. Jitter is the variation in arrival times for packets between two participating endpoints

Defining network metrics that matter to an organization depend on the use case.  If someone wants to perform the task of bulk data movement, often it is desirable to have a path high throughput.  This implies other desirable features:

  • Low (or zero) packet loss
  • Stable jitter
  • low path utilization

The perfSONAR tools are able to measure some metrics, and can be viewed together to provide a picture of what is capable on a network. 

perfSONAR Tool Overview

The following tools are listed, along with metrics they measure:

    • The OWAMP tool is designed to measure one way latency, packet loss, packet duplication, and jitter on an end-to-end basis.  Small UDP packets are used for this measurement, along with a TCP control channel for test negotiation. 
  • BWCTL, Iperf, Iperf3, Nuttcp
    • BWCTL, and the tools it calls (iperf, iperf3, and nuttcp) measure achievable bandwidth (throughput) between hosts on an end-to-end basis.  These tools rely on the hosts being properly tuned for their representation of the underlying network.  Some of these tools can also record TCP statistics such as retransmissions, or UDP performance such as packet loss, jitter, and duplication.  
  • Traceroute, Tracepath
    • Both tools measure the network layer routing (e.g. OSI Stack Level 3) between points using ICMP packets.  Options allow the user to use other packets (UDP, TCP).  Tracepath can also record statistics on MTU. 
  • Ping
    • Ping is a tool used to record round trip delay, and can also record related statistics for a train of packets such as loss and duplication. 
  • SNMP
    • SNMP is a method for receiving passive counter values from networking devices.  For example, each router interface has a notion of when it had to drop a packet, if the packet had an error, or the amount of packets that have flowed through since it started counting.  Using tools that regularly ask for this data, one can generate graphs such as network utilization. 

Survey of Internet Performance Measurement Platforms

Vaibhav Bajpai of the Computer Networks and Distributed Systems (CNDS) Lab at Jacobs University Bremen in Germany prepared a survey of performance measurement platforms and standardization efforts.  The article's abstract is as follows:

A number of Internet measurement platforms have emerged in the last few years. These platforms have deployed thousands of probes at strategic locations within access and backbone networks and behind residential gateways. In this paper we provide a taxonomy of these measurement platforms on the basis of their deployment use-case. We describe these platforms in detail by exploring their coverage, scale, lifetime, deployed metrics and measurement tools, architecture and overall research impact. We conclude the survey by describing current standardization efforts to make large-scale performance measurement platforms interoperable.

This work can be found using the following links:

IEEE Library:
Mirrored Copy: