About Latency Claims

Recently we have seen a small flood of articles and promotional materials that make claims about network latency. We find many of those claims to be hyperbolic and sometimes quite misleading. Let us begin, as we almost always will begin, by informally defining the terms "latency" and "jitter".

Latency: (Also "Delay") The time required for a packet to travel across a network from its source to its destination. (Sometimes latency is described as the round-trip time; care must be taken to know whether the term is being used to describe one-way or round-trip.)

Jitter: A mathematical measure of how much latency varies from packet to packet. There are several mathematical formulas for calculating jitter. Again, one must be careful to know whether one-way or round-trip times are being measured. Being a mathematical expression, a value of N milliseconds of jitter does not mean that no packets will experience latency variations outside of a N millisecond window, rather that such variances will tend to be infrequent.

A far more detailed definition can be found in the IETF's RFC 3393 - IP Packet Delay Variation Metric for IP Performance Metrics (IPPM).

All networks exhibit latency.

All networks must obey the laws of physics, especially the speed of light. There is no such thing as zero network latency.

Admiral Grace Hopper had an amusing, and illuminating, way of describing network delay.

For some purposes low latency may be the practical equivalent of zero latency. Almost all networks exhibit some degree of jitter. However, with enough tuning, traffic engineering, and often an over-provisioning of network resources, jitter can be limited. Sometimes this comes at the cost of increased latency.

Latency and Remote Operation of Robots and Vehicles

Below is a link to an interesting short piece describing the elements, and the latency contribution of each element, of a remote-control system for automobiles and robots:

Also see:

Some people, especially people who do high-speed trading of financial instruments, are willing to invest a great deal of money to reduce network delays. The topic has even been raised on TV shows and has been the subject of at least one dramatic film.

  • The Hummingbird Project - A movie on the subject of network latency

5G

The latest generation of mobile/wireless networks is "5G" from the 3GPP (One must take care to distinguish the core 5G proposals from add-ons often called "millimeter wave".)

5G does indeed improve latency as compared to its predecessor. This will be a boon for direct, non-intermediated communications between vehicles and other vehicles and roadway infrastructure (e.g. traffic signals.)

However one must understand that in most other contexts a 5G network path may be only a small piece of an overall networking system.

An analogy may be drawn: Your overall commute between home and work may involve neighborhood roads, highways, trains, buses, and perhaps bicycles. A 5G link is like your driveway — it is only a tiny part of the entire trip. Widening your driveway to reduce the time it takes to get out of your garage might be nice, but it makes very little difference in the context of the entire trip.

In other words, improvements given by a 5G hop may be drowned out by the far greater sources of latency and jitter formed by the rest of the end-to-end network path.

Let's look at two claims being made:

Remote-Control Surgery Over 5G

This video describes doctors who are 50km from the actual surgical site.

The video says that the latency is 100ms (but does not say whether that is one-way or round-trip.)

Anyone who has tried to speak or operate equipment with a delay knows that human reactions can get very odd - we all routinely experience these effects when we chat over a bad mobile phone connection or try to perform a group sing-a-long of "Happy Birthday" over a Zoom conference.

Those who operate devices (such as a pilotless aircraft or drone) over long delay paths often require special training. It would be interesting to learn how the surgeons in the video adapted to the delay. It would also be interesting to understand how this latency would impact their ability to react should something unexpected happen, such as a dropped or broken tool.

It is interesting to note that the ITU recommends for voice that the mouth-to-ear (one-way) delay not exceed 150 milliseconds:

SERIES G: TRANSMISSION SYSTEMS AND MEDIA, DIGITAL SYSTEMS AND NETWORKS - International telephone connections and circuits – General Recommendations on the transmission quality for an entire international telephone connection (See Appendix II, page 12.)

So even at the short distance shown in this video - 50km - the delay is already approaching the limit that the ITU recommends for voice. One can only imagine what remote-control surgery would be like over greater distances or over the very long delay of a high-orbit or geosynchronous satellite.

Remote Operation of Vehicles Over 5G

The authors were both involved with the DARPA Robotics Challenge. One of our tasks was to introduce communications delays (and other troubles) into the link between the robots and their operators.

The results were often amusing. In the case of vehicles and 5G one must distinguish between two cases:

  • Direct, non-intermediated interactions between vehicles (or between vehicles and highway infrastructure or other nearby things, such as 5G equipped pedestrians or bicycles.)

  • Indirect, intermediate interactions.

An intermediated communication is a network connection that flows through one or more relay points, such as a mobile phone tower. If there is only one, fast intermediary device - such as a repeater on a nearby streetlight then the delays can approach those of the non-intermediated case.

However, one can adopt a very crude rule of thumb that intermediated interactions can be, and often will be, an order of tens magnitude (or more) longer than non-intermediated interactions.

But even direct interactions will have many elements that contribute to the overall latency. See: Network Considerations for Remote Vehicle Operation

For many vehicle interactions, such as with other nearby vehicles or road infrastructure, interactions with 5G will be direct and fast. That is very good.

However, remote operation will almost necessarily be over an intermediated pathway. That means that the entire system will have to be carefully tuned and, in many cases, speeds limited to confine the distance a vehicle could move during the time needed for the remote operation sense-and-control loop to operate.