Phil Hippensteel on The Effect of Network Packet Loss2/08/2017 11:21 AM Eastern
In our last newsletter, we investigated the effect of network latency on IP video. We found that if latency was high, video transported with TCP might pause but would show no other signs of deterioration. On the other hand, when UDP was used as the transport protocol, buffers in the network or at the receiver might empty causing the video to pause. Or, buffers in the network could overfill, causing packets to be dropped. The result would be an outcome commonly called tiling. When this happens, small blocks or slices of the screen are severely distorted or appear black.
Our topic today pertains to what happens when the network drops packets that could be the result of latency or other factors. These other factors could be signaling errors with wireless transmission, signal distortion due to poor cabling, or other forms of signal interference. Packets can also intentionally be dropped by switches and routers as a mechanism to control network congestion. Again, we find the degradation of the video will depend on whether the transport protocol is TCP or UDP.
When TCP is being used, packets that are dropped will eventually be retransmitted. However, it is a little known fact that such retransmissions can happen after dozens or even hundreds of succeeding packets have be sent. This occurs because modern networks use very large buffers that store many hundreds of packets. The result of packet drops is usually a slowing of the rate that TCP sends packets into the network. Sometimes the sending rate is cut by half or more. As a result, the playout buffer at the receiving end can empty and there could be a pause in playing the video. Note, that when the video is played, it should show no signs of degradation because all dropped packets would have been received.
UDP carried video is much different. First, if you recall the newsletter in which we discussed the mpeg transport format, you will recall that video frames are compressed as I, P, or B frames. I-frames are the largest and are essentially jpeg images. P and B frames are compressed using the information in the I-frame and are significantly smaller. Suppose the video being transported is 720p30. An I-frame will be a representation of what has been captured by the camera in 1/30 of a second. In nearly all cases, the data for that I-frame will be transported in hundreds of consecutive IP packets. If one or more IP packets are dropped, the data representing a small block or slice of the picture will be distorted or it could possibly be black. Even worse, the I-frame data is the basis for computing the succeeding P and B frames. So, the distortion might be continued until the next I-frame is computed. This could be as long as a half second or more.
So, to summarize, TCP video such as adaptive bit rate and progressive downloads might pause when confronted with packet loss. However, TCP retransmission will mean the video will eventually play as intended without distortion. UDP video such as IPTV, digital signage and video conferencing will typically exhibit tiling, distortions of blocks or slices in the viewed video frames.
In our next newsletter, we will look at the impact of network jitter on IP video.