About network End-to-end path performance testing

In recent years, VoIP, VOD and other real-time applications in the network application of a large number of increase in real-time applications require the network to provide service assurance or differentiated services, network path delay characteristics and transmission rate more and more people pay attention to. Network latency and effective bandwidth directly reflect the performance of the path and the level of service levels provided to the end-host. But they are all variables that change with time randomly, and the dynamic change of these path characteristic parameters can obtain the resource usage and performance trend, and serve as the basis of network path performance evaluation and real time end-to-end performance management.

First, End-to-end path performance issues

The goal of End-to-end performance management is to provide a stable and usable connection with a certain service guarantee for the user. Its scope has exceeded the flow engineering and capacity planning and other traditional issues, including fault detection, Routing and configuration Management, traffic analysis and so on a wide range of content. For the user, the delay of the path and the effective bandwidth of the path determine the service level, that is, whether a new connection can be initiated, a certain resource is guaranteed in the local node, and an acceptable response time is reached. For ISPs, it is also concerned with properties and parameters such as the long-term utilization of network paths, routing accessibility, bottleneck bandwidth, and other end-to-end performance. These parameters are generally more stable, in which connectivity and path stability reflect the overall performance of the path, while the path asymmetry affects our measurement of network latency.

of connectivity

Connectivity refers to whether the hosts in an IP network can reach each other, or whether the underlying connections between one or more nodes are available. Connectivity describes the reliability of the network and is the basic condition to complete the network business. The ping program can be used to determine whether a remote device on the network is up to the local system and to help debug connectivity issues between systems. We can run a ping from the measurement host to the destination of the network path, issue an ICMP request and wait for a valid ICMP response to determine the availability of the network connection, and can isolate the solution by the type of ICMP answer by choice and many other network failures or problems. Therefore, connectivity is the basis to ensure the stability of path performance.

Path MTU

The MTU is the maximum amount of data in a network connection that allows a physical frame to be transmitted, and the TCP/IP driver needs to know the value to set the appropriate packet length to avoid unnecessary splitting processes. The MTU defaults for different networks are not the same, and the standard value for Ethernet is 1500, while most ISPs provide a standard value of 576 for dial-up networking. Theoretically, as long as the network connection stability is high, the larger the MTU is the higher the transmission efficiency. However, if the router's MTU is small on the network or transmission path of the receiving machine, the data frames will have to be split again, which reduces efficiency. Therefore, the optimized MTU should be equal to the minimum MTU value (that is, the path MTU) on the entire transmission path. RFC1191 describes the discovery mechanism of the path MTU, sets the not Fragment flag field in the group header by sending packets and determines that the ICMP error message implementation is returned. The length of the first packet sent is equal to the exit MTU, and each time an ICMP cannot be fragmented, the packet length is reduced, with one of the following minimum MTU values sent. Because the number of MTU values is always limited, it is always possible to find the smallest MTU in the path. So detecting the Path MTU and setting the MTU value of the system can reduce the segmentation of IP packets and improve the transmission efficiency of the path.

Routing asymmetry

Asymmetric routing, which is inconsistent with the path of a certain node, is different in two directions, thus causing many problems in network measurement, especially based on the source-side measurement. It is still very difficult to detect the asymmetry of routes.

Path stability

The stability of End-to-end path is a key parameter to measure network performance, and path stability can be divided into path stability and time stability. The stability of spatial path is determined by the size and mode of the route, and the unstable route may be caused by the ill routing behavior, such as routing asymmetry, cyclic routing, routing oscillation, or frequent interruption of connection. Incidence (prevalence) and persistence (persistence), respectively, define stability from the frequency and invariance aspects of a particular route. The time stability of the path emphasizes the relationship between path performance and time change. In this paper, the evaluation index of path stability is given based on the delay characteristic and transmission rate of the path, and it is used as the basis for the evaluation of Network path performance and the real-time End-to-end performance management.

Second, performance indicators

Path delay stability parameter

The delay variable of the measurement is X, and the network delay measured by T is the time delay change in the interval. If the sample value for each measurement in the time period is, then the sample variance of the set of delay samples is, of which =. Suppose a network latency jitter target threshold defined for SLA (Service level agreement) or some application of quality of service requirements. According to the jitter degree of the time delay, the delay jitter bounded proportional coefficient is defined as: generally preferable. The delay jitter bounded proportional coefficient reflects the proportion of delay jitter between adjacent two measurements, generally set at that time, the connection has a good stability, the greater, that is, the closer to 1, the more stable the connection. But the delay jitter bounded proportional coefficient can not reflect the maximum delay jitter, because the queue delay with the background flow and the larger fluctuations, the definition of the following form of the average jitter/maximum jitter coefficient: Set, think the connection has a good stability, the larger, the more stable the connection. Combined with the above two parameters, the path is determined to be stable for delay when the path satisfies and simultaneously.

Path Rate Parameters

The path delay stability parameter reflects the stability of the whole link operation, but it does not reflect the difference of the link rate. A certain link may be delayed, but the delay jitter is small, and the loss rate is very low, then the path delay stability parameter is difficult to reflect the actual transmission quality of the link. Through the collection of SNMP network device MIB Interface group data, we can compute the interface data transfer rate V: The total number of bytes received and sent by the inner interface respectively. By continuously collecting the flow data of multiple interfaces in the path, the minimal interface transfer rate is notified to the user as the path rate parameter to reflect the actual transmission performance of the current path.

Data acquisition

We select the measurement host of the laboratory to the Hunan University DNS server as the experiment object, the path passes through two routers, respectively through 100M Ethernet and 1000M Ethernet link. We used ping to collect the time delay data for each 10ms measurement for nearly two days, and obtained the interface rate data per 5ms by reading the first Router MIB library.

Third, end-to-end path performance measurement

We use the above measurement methods for the following reasons. In general, the measurement of network performance parameters takes the active measurement method, that is, on the specified path, the detection flow data are sent according to certain rules to measure whether a host on the network is up to (reachability), network delay, packet loss in network, network throughput and other link performance parameters or measure the hop number of network path. Routing jitter, routing symmetry, and so on, routing information from the source end to the destination. The active measurement will introduce more measurement flow in the network, which will have a certain effect on the actual performance of the network. (☆ Introduction to Programming Network ☆)

Because the measurement of path bandwidth needs to send a large number of packet data to the network, it consumes a lot of effective bandwidth, which makes it very damaging to the user application and the network to measure the bandwidth in an active way. SNMP can be used to query the network device interface of the flow rate data, combined with SNMP and network round-trip delay measurement evaluation path performance, the measurement sent to the lowest number of packets.

Analysis of delay distribution data

The distribution of the time delay data can be seen that the measured value of the RTT is very unstable, the delay time is up to 25.198ms, the minimum is 2.091ms, and it is close to the intrinsic delay. But most of the scattered points in the 10ms up and down 5ms range, only very few points in greater than 20ms less than 5ms range, indicating that the path is relatively stable. The average jitter/maximum jitter coefficient is calculated with the delay jitter bounded proportional coefficient, which is in accordance with the actual network operation condition.

Analysis of delay jitter data

In 40 hours, the jitter distribution of delay jitter is measured at 10 minutes, and when the critical value of delay jitter is =4.629ms, the 16ms boundary of Jitter is at worst, far exceeding the expected value. This means that the path is unlikely to achieve good performance for real-time applications.

Link Interface Rate Analysis

At the same time, the data of router MIB interface traffic rate can be seen that the utilization rate of this path is very low, especially the outflow speed is very small. In practice, we can increase the utilization of the network by increasing the load, and compute the maximum data transfer rate that the link interface can achieve.