Buy Cisco Stock...Then Wear Stripes?Buy Cisco Stock...Then Wear Stripes?
In my previous blog about Telepresence, Presence and "Presence," I cited a talk that a gentleman named Anthony Townsend delivered to a group called the Institute for the Future. In the course of that talk, Mr. Townsend dropped this interesting little tidbit of information:
September 9, 2008
In my previous blog about Telepresence, Presence and "Presence," I cited a talk that a gentleman named Anthony Townsend delivered to a group called the Institute for the Future. In the course of that talk, Mr. Townsend dropped this interesting little tidbit of information:
In my previous blog about Telepresence, Presence and "Presence," I cited a talk that a gentleman named Anthony Townsend delivered to a group called the Institute for the Future. In the course of that talk, Mr. Townsend dropped this interesting little tidbit of information:
The bandwidth requirements [for telepresence] aren't great (about 15 Mbps for Cisco's system), but there are some funny impacts of how you dress--for instance, if you wear a striped shirt, you completely confound the compression algorithms and dramatically increase bandwidth requirements.
I asked myself: Can that possibly be true? Then I asked John Bartlett the same question, and John sent me this response:
His first comment is that the bandwidth requirements aren't great (15 Mbps), and I would say this is relative. Many network folks would have a real headache with an application that demands 15 Mbps continuously.
The striped shirt example is a partial truth. My greatest experience is with the Polycom codecs. Its quite possible that the Cisco codecs do something weird. Here is how I think about this:
The compression algorithm used by the codecs is incremental. This means that it first sends a whole picture image. Then in the next frame, it only sends the differences from the previous frame. So if you are looking at a conference room with folks sitting at a table, much of the image (table, wall, pictures on the wall, etc) has not changed. Small portions need to be updated such as faces, eyes, hands. Thus the codec has little information to send, and so it can send with improved resolution or in some cases lower the bandwidth being used.
A striped shirt has the effect of causing a lot of change in the image. A flat blue shirt that moves ¼ inch to the right will still leave large fields of essentially the same color, and cause very little inter-frame updates. But a striped shirt moved ¼ inch to the right will cause a lot of updates because all the white parts just turned colored, and vice versa. The constant changes that this causes can cause the codec to work hard to process what would otherwise be a simple image.
Given this scenario, the Polycom codecs will push as much information across the link as they can but will still stay within the prescribed bandwidth limit. If the call was set up at 4 Mbps/codec (typical of a Telepresence system) it will max out at this number. If the scene settles down (or that guy changes his shirt) the bandwidth may drop below 4 Mbps. Most video vendors who have been in the business awhile will have this style of codec, because in the past these codecs were required to work within a fixed limit bandwidth as prescribed by ISDN. If you are on a 384K call on ISDN, 384K is all you get.
Cisco may have optimized their codecs to temporarily take advantage of the fact that they are on an IP network that has bandwidth headroom. Some of their literature implies that the routers should allow short bursts of additional traffic. I don't have direct experience with their devices to know how much of a burst this situation might cause. But soon they would have to settle down again or they too will exceed the QoS bandwidth of the IP network and packets will start to be dropped.
The striped shirt example is a partial truth. My greatest experience is with the Polycom codecs. Its quite possible that the Cisco codecs do something weird. Here is how I think about this:
The compression algorithm used by the codecs is incremental. This means that it first sends a whole picture image. Then in the next frame, it only sends the differences from the previous frame. So if you are looking at a conference room with folks sitting at a table, much of the image (table, wall, pictures on the wall, etc) has not changed. Small portions need to be updated such as faces, eyes, hands. Thus the codec has little information to send, and so it can send with improved resolution or in some cases lower the bandwidth being used.
A striped shirt has the effect of causing a lot of change in the image. A flat blue shirt that moves ¼ inch to the right will still leave large fields of essentially the same color, and cause very little inter-frame updates. But a striped shirt moved ¼ inch to the right will cause a lot of updates because all the white parts just turned colored, and vice versa. The constant changes that this causes can cause the codec to work hard to process what would otherwise be a simple image.
Given this scenario, the Polycom codecs will push as much information across the link as they can but will still stay within the prescribed bandwidth limit. If the call was set up at 4 Mbps/codec (typical of a Telepresence system) it will max out at this number. If the scene settles down (or that guy changes his shirt) the bandwidth may drop below 4 Mbps. Most video vendors who have been in the business awhile will have this style of codec, because in the past these codecs were required to work within a fixed limit bandwidth as prescribed by ISDN. If you are on a 384K call on ISDN, 384K is all you get.
Cisco may have optimized their codecs to temporarily take advantage of the fact that they are on an IP network that has bandwidth headroom. Some of their literature implies that the routers should allow short bursts of additional traffic. I don't have direct experience with their devices to know how much of a burst this situation might cause. But soon they would have to settle down again or they too will exceed the QoS bandwidth of the IP network and packets will start to be dropped.
