Network Latency Milliseconds Per Mile

The longer the fiber-optic cable, the greater the latency.

Latency is the time lag between the moment you send data across a network and when it arrives. Latency is measured in terms of milliseconds per mile. Although the lag in most cases is only a fraction of a second, that can be significant for some real-time applications. There are ways to cut a network's latency.

Sources of Latency

In a vacuum, signals between computers travel at the speed of light, 186,000 miles per second. In a fiber-optic cable, they slow down closer to 122,000 miles per second. The loss of speed is roughly 8.2 microseconds per mile, or .82 milliseconds per 100 miles. If the data packet has to pass through a router or a switch, or your network uses NAT -- network address translation, a system for sending packets for your network to your router's address -- the latency increases.

Significance

Network latency matters more with small packages than big chunks of data. On big, slow-moving data packets, the added drag of a millisecond or two is hardly noticeable. With small data packets that should move swiftly, the added time can be significant. High latency is particularly noticeable in real-time voice or video communications -- a question may not be followed by an immediate answer.

Distance

One way to reduce latency is to reduce the miles that a signal has to cross. A single mile of cable produces .5 percent of the delay a 200 mile stretch of cable does. If you're planning a new location for an office that will become part of a wide area network, try to minimize the distance from the next node or the hub of the network. If that isn't an option, you'll have to consider alternative solutions.

Solutions

If your switch uses hardware-assisted forwarding -- a method for steering packets to the right address -- it cuts latency considerably: Anticipate an added delay of 25 microseconds going through the switch, much less than other switches would cause. If you have heavy congestion on your network -- packets coming into a router faster than they can leave -- cutting the congestion will cut latency. Adding hardware to allow parallel processing, where the network handles several jobs at once, also helps.

References

• Cisco: Design Best Practices for Latency Optimization
• Stanford University: It's the Latency, Stupid!
• SQSkills.com: Importance of Network Latency When Using Data Mirroring

Resources

• University of Virginia: Networks, the Internet and the World Wide Web