Enterprise Technology By Phil Edholm

One topic that continues to intrigue me is the relationship between the demand for bandwidth created by the applications we run and the bandwidth available in the networks we use. When these are out of alignment, the system is unbalanced and can create significant operational and user issues.

This thought process led to the Edholm's Law of Bandwidth 4 years ago and thoughts on how to manage delivery of content to different networks that a device might be connected to. You can click on the link to see the article published by the IEEE in 2004. I need to write a quick update to the Law of Bandwidth at some point, and I will write a future post on content management.

However in this post I wanted to discuss the drivers of bandwidth to the user. The included chart shows what I believe are the three drivers of user bandwidth requirements; Moore's Law, Metcalf's Law, and the changing type of interfaces.

Moore's Law is a well understood and says that things that are based purely in silicon double every 18 months (processors, memory, etc.). This is due to a trace on a chip shrinking 30% in the manufacturing process and the number of devices is the area (.7 x .7 = .49). It would seem that there is a corollary that the size of things (software, applications, files, etc.) operated on these devices follows Moore's Law (PowerPoint charts double every 18 months). So, if we extend this to other types of traffic (audio, video, etc.) we can also assume that they grow as the capacity to display present enables higher quality, resulting in increasing size (either file or streaming).

Metcalf's Law says the value of a network is an exponent of the number of users. While I do not want to debate the actual exponent value in this post (there has been a lot of dialog on this subject recently), I believe there is a corollary also to Metcalf; user communications bandwidth grows at some exponent of the growth in users. In other words, a new user does not only communicate with one user, but with many. Email cc lists are a great example of this, while conferencing and the existence of continual conferencing may be another.

Finally, advances in computer to human I/O change the paradigm and over a relatively short period can generate order of magnitude increases in bandwidth requirements. The change from text to graphics generated a one to two order of magnitude increase in bandwidth. A text screen is typically 400 bytes, while a graphic screen can be 50-100Kbytes, an increase of 10-100x. Similarly, while a screen is large, the transition to audio, both file/streaming and voice interactive, changed the bandwidth once again. Wile the stream is relatively low in instantaneous bandwidth, it is relatively continuous. That VoIP call at 64 Kbits/second (assume no compression) is 8KbBytes/second or 480 KBytes/minute or 28,800 KBytes per hour. Assuming a browser screen is 50 KBytes, then a 1 hour conference call is equivalent to 576 screens. Assuming that the user browses for 20 minutes during the hour and sees 3 screens per minute, that is 60 screens, so the audio is 10x the browsing. Finally, video ups the ante once again. with frame rates of 24-30 Fps and screen sizes up to the 1920x1080 of HD, the bandwidth explodes. An uncompressed HD at 1080P and 24 bits of color (8 bits per color for the 3 primaries) is 178 Mbits/second. obviously compression allows this to be dramatically reduced, but it is still another order plus of magnitude.

So, while bandwidth growth may seem at some level to be linear, it is really driven by these three factors and is therefore both linear and abrupt as new technologies enable changes in the I/O operation that has dramatic impact on the bandwidth required.

In a future blog I will talk about whether this trend will end and why, but for now it is interesting to observe that bandwidth demands continue to increase exponentially. The challenge is to build networks that can deliver the bandwidth needed to meet customer expectations of the teh performance of an application at a specific point in time. In some ways this is the "designing the network for today's problem", when over the lifespan of the investment a major change may occur. In many ways, bandwidth growth is the natural cousin of Hyerconnectivity and as connectivity explodes so will bandwidth per device.