In a previous edition of this column, “placid” video streams were defined as having a smooth, constant flow that have neither bandwidth peaks nor deficits, but are ideally suited for constant bit-rate video signals. Verifying that a stream is indeed placid requires a measurement technique, of which several are commonly used in the industry. The difficulties in using Peak/Average Ratio (PAR) were discussed in the previous column, because of the sensitivity of the measurement to the size of the sampling window employed.
In this column, two other measurement techniques will be examined. The first, Packet Delay Variation (PDV), can be used to measure the performance of a network that is used to deliver packets between two locations. The second, Media Delivery Index (MDI), can be used to characterize a stream at multiple points along its journey from a source to a destination. These techniques should be used for different purposes.
PACKET DELAY VARIATION
Even if a source generates a perfectly placid stream, there are multiple points in a network where packets can be delayed or even discarded during heavy traffic loads. IP routers and switches use queues to manage how packets are forwarded along their route; the amount of time a packet spends in a queue is influenced by its priority level and its order of arrival in the queue.
If a switch or router happens to receive multiple packets at the same time at different inputs that are destined for the same output, then the device will need to choose the order in which to send out the packets, and temporarily store packets that cannot be sent immediately. These short periods of storage, along with other network transmission artifacts, can cause variations in the delay across the network.
As defined in IETF RFC 3393, measuring PDV requires at least two packets to be sent across two measurement points in a network. First, a measurement is made of the amount of time the first packet takes to travel between the two points in the network (i.e. the transit delay). Then, this measurement is repeated for the second packet. The packet delay variation is then calculated by subtracting the first transit delay from the second to determine the PDV. Note that this value can be positive or negative, or even zero if there is no delay variation present on the circuit.
PDV can be used as an indicator of network performance, because it shows if there are any places along the transmission path where the packets experience variable amounts of delay due to buffering, congestion or other impairments. Unfortunately, PDV is not well-suited for characterizing individual points along a network route or measuring the behavior of packet sources because PDV is intrinsically a measurement that is made between two points.
MEDIA DELIVERY INDEX
Fig. 1: MDI streams
The core concept of making an MDI measurement as defined in IETF RFC 4445 is to determine the size of the buffer that would be needed to compensate for any unevenness in a packet flow. Smaller required buffers indicate smoother, more placid streams, with the minimum possible buffer equal to one packet full of data divided by the nominal media rate. This measurement is called the MDI:DF for Delay Factor, which indicates the amount of delay (in milliseconds) that a perfect receiver would need to introduce in order to fully buffer the variations in the incoming signal.
Using the example discussed in the January column, a 200 Mbps stream being carried as a sequence of perfectly spaced 10,000 bit packets (1,250 bytes) would have an MDI:DF of 50 microseconds (10,000 bits divided by 200 Mbps).
Calculating the MDI for a packet stream requires making two measurements when a new packet of data arrives. One measurement is the size of the Virtual Buffer (VB) just before the packet arrives. The second is the VB size just after it arrives. These measurements are typically accumulated over a one-second measurement interval.
All of the VB measurements made during the time interval are then compared, and the DF is calculated as the difference between the highest VB value and the lowest VB value. In Fig. 1, a perfectly placid stream is shown in A; while B shows how early packet arrival times increase the maximum VB level and how late packets decrease the minimum VB level. In both cases, the amount of buffering required at a receiver grows and the measured value of MDI:DF increases.
The primary benefit of using MDI:DF is that it can be used at any point (or multiple points) within a network, thereby allowing individual devices to be measured and certified. This is particularly beneficial for specifying network components, such as signal sources or receivers, since their generated MDI:DF (or their input tolerance for various levels of MDI:DF) can be qualified on a standalone basis. Another benefit of MDI:DF is that it can be used to measure variable bitrate streams, which can be a challenge for other measurement techniques.
Thanks to Jim Welch of IneoQuest (author of RFC 4445) for his helpful comments.
Wes Simpson is an independent consultant and educator for IP and video technology. He will teach a pre-conference class on OTT Video at VidTrans 2015 in Marina del Rey, Calif. on Feb. 23.