Overview of Flow Recognition

Flow recognition enables a device to collect the septuple information (source and destination MAC addresses, source and destination IP addresses, source and destination port numbers, and protocol type) of the upstream traffic on an interface, and identifies a flow based on the information. The device then sends the flow's statistics (such as the numbers of packets and bytes, timestamps, and interface information) to a controller for display and traffic identification.

Background

In the radio and television industry, especially in TV stations or media centers, IP-based production and broadcasting networks are gaining in popularity. Related IP standards are being formulated, which is an important step in the development of the 4K industry. However, IP-based production and broadcasting networks require switching among multiple video sources or cameras during video production and live transmission to achieve the optimal video display effect. Currently, IP-based devices use IGMP to switch between different multicast groups. When IGMP starts or stops multicast forwarding, it does not determine the frame boundary of a video. As a result, the forwarded content is incomplete and the video is damaged (such as artifacts, jitters, black screens, and static frames).

The quintuple information (source and destination IP addresses, source and destination port numbers, and protocol type) may not distinguish between flows. To resolve this issue, configure flow recognition based on the septuple information (source and destination MAC addresses, source and destination IP addresses, source and destination port numbers, and protocol type) on a device. A controller calculates the flow rate based on the information reported by the device, and determines whether the flow is video or audio flow based on the flow rate. The device then replicates and broadcasts the identified flow quickly and accurately.

Flow recognition is used to identify video and audio flows on IP-based production and broadcasting networks.

Implementation

On the topology shown in Figure 1, traffic enters the device through the inbound interface. The device extracts the septuple information from the traffic and generates matching rules based on the septuple information. Each flow that matches the septuple information has a statistical ID. The device collects statistics about the numbers of packets and bytes for each statistical ID. The device sends the statistics of each statistical ID to the controller over telemetry. The controller calculates the flow rate based on the current and last data records to identify the video or audio flow.

Figure 1 Flow recognition networking

To calculate the flow rate, the controller needs to receive the data information of each flow. Table 1 describes the flow data fields collected and sent by the device to the controller.

**Table 1** Flow data fields
Field	Type	Description
srcMac	Character string	Source MAC address
dstMac	Character string	Destination MAC address
srcIpAddr	Character string	Source IP address.
dstIpAddr	Character string	Destination IP address
srcPort	Integer (4 bytes)	Source port number
dstPort	Integer (4 bytes)	Destination port number
protocol	Integer (4 bytes)	Protocol type
direction	Integer (4 bytes)	Direction in which the traffic enters the interface: 0: inbound 1: outbound
ifName	Integer (4 bytes)	Interface name
timeStampSec	Integer (4 bytes)	Timestamp in seconds, which is used to calculate a time difference
timeStampNsec	Integer (4 bytes)	Timestamp in nanoseconds, which is used to calculate a time difference
packetNum	Integer (8 bytes)	Number of packets
bytesNum	Integer (8 bytes)	Number of bytes of a packet

If the device consecutively collects statistics about a flow twice, the flow's rate is calculated as follows:

Number of packets forwarded per second = (packetNum2 - packetNum1)/(timeStampSec2 - timeStampSec1)
Number of bytes forwarded per second = (bytesNum2 - bytesNum1)/(timeStampSec2 - timeStampSec1)