Browse Prior Art Database

Use of a Programmable Network Processor to Measure Packet Flow Characteristics Using Precision Packet Insertion and De-insertion Disclosure Number: IPCOM000015946D
Original Publication Date: 2002-May-28
Included in the Prior Art Database: 2003-Jun-21

Publishing Venue



For precise measurement of traffic flow characteristics, one needs to measure the latency of a packet across a network. Often this measurement needs to be repeated multiple times to obtain an "averaged" result and to obtain jitter and other information. With the advent of high speed networks, this becomes difficult. The network processor can be used to precisely insert "marked" packets into a data stream and to remove them from a data stream, resulting in highly accurate network characteristic measurements in a very high-speed environment. Figure 1, illustrates this with a network of Network Processor Systems, 3 of which are measuring the latency and jitter characteristics of two given flows. By using the functions of a Network Processor, a "marker packet" can be precisely inserted by a Network Processor into a flow at the beginning observation node and removed by the network processor at the ending observation node (Figure 1). By use of time stamps, the transit time of the packet and the interpacket arrival time can be obtained. Given this information, for example, latency and jitter for a particular virtual path or flow can be derived. Note that either or both a path (all the packets leaving a port) or a flow (a subset of packets leaving a port that have been identified as part of a logical flow) can be measured. As Figure 2 illustrates, the beginning observation node precisely inserts (e.g., after the Nth packet) a "marker packet" into the data stream. A "marker packet" is a packet generated by the NP acting as the beginning observation node that contains a time stamp of when the packet was enqueued, a correlator that associates the packet with a particular measurement, and an address or marker that allows identification of the "marker packet" by a ending observation node. The endpoint can be identified in one off three methods, the NP IP address itself,