Ping the call test to test the end-to-end bandwidth of the data network.

With the development of various data services of China Mobile, more and more attention has been paid to service quality assurance for end-to-end users. Many services must be guaranteed with sufficient end-to-end available bandwidth. For carriers, the basis of core operation network optimization is end-to-end bandwidth measurement. The measurement of the current actual bandwidth not only has a certain significance for evaluating the quality of network engineering, but also affects network resizing and resource utilization. In the market, existing network management products have a single function in Bandwidth Measurement and are expensive. They are not suitable for centralized network management. To meet this requirement, the ping latency dialing test module has been deployed in China Mobile's integrated data network management phase II project. This module is mainly used to test whether data devices and business hosts are online, obtain the end-to-end latency.

In this case, Shanxi Mobile uses the existing network management module to propose a method to test the end-to-end bandwidth of the data network by using Ping call tests. This method applies the variable packet length Detection Method in the traffic engineering to realize the bandwidth estimation of each hop on the end-to-end path. It is worth mentioning that this test method can be achieved only after a small number of modifications are made to the call test module of the network management system. Therefore, the cost-effectiveness is very superior.

1. Improve the detection method

1.1 variable packet length Detection Method

The variable packet length detection method uses the ping test results to test the capacity bandwidth of each hop on the path. The network administrator can directly read the port information of the vro or vswitch to obtain such information. However, the network capacity of the end-to-end network and the network capacity that is not under management are unknown, in addition, in the actual network environment, the theoretical port information differs greatly from the actual bandwidth capacity. In addition, end users and service providers can only estimate the bandwidth and cannot directly obtain network capacity parameters. For problems such as engineering line quality and network bottlenecks, operators can only find out through bandwidth estimation, which is inconvenient in the actual network environment.

Variable Packet Length detection is mainly used to send packets of different sizes from the source to any device in the path. By calculating the function relationship between the round-trip latency and the package size, you can obtain the bandwidth capacity of each segment of the path. Specifically, the TTL domain of the IP header is used first (same as the common tracerouter tool) to force the packet to time out in a specific hop. The router of this hop will discard the test packet and send the packet back to the source device using the timeout error message of the Internet Control Message Protocol (ICMP. The source device obtains the path device information, and then sends the ping test packet to calculate the round-trip latency of this hop through the received ICMP packet.

Round-trip latency consists of three parts. The first part is the latency when a packet is sent to the line. This time-sharing delay is related to the packet size. If the packet length is L and the bandwidth is C, the transmission latency is t = L/C. The second part is the signal transmission latency, which is basically at the speed of light level in order of magnitude. It is irrelevant to the packet size, as long as the path is not changed, this delay will not change. The third part is the queuing delay, which is mainly related to the cache of the vro switch. Due to the large number of packets in the network, the delay may be queued for sending in the cache of A vro, this will lead to queue latency.

The variable packet length detection method is to send many fixed-size packets to each layer-3 device in the path from the source device node. This method assumes that at least one packet can avoid queuing delay, the round-trip latency certainly does not include the queue latency. Therefore, we can think that the packet with the minimum round-trip latency only contains the transmission latency and transmission latency. When we set the round-trip delay of the I-hop minimum to Ti (l) when sending an L-Length Packet ):

(1)

Here, CK is the capacity of the K hop, latency is the latency unrelated to the package size, and β I is the slope of the Gini function ,. So there are: (2)

This is the mathematical expression of the Variable Packet Length detection method, that is, by finding the slope of the round-trip latency function of the two devices where a certain hop is located, the capacity bandwidth of the hop is calculated from the slope.

Based on the transformation of the ping call test module, we tested the current network. Expected test result 1 is displayed. This test calls a layer-3 switch in the network. The packet size increases from 100 bytes to 1400 bytes (note that the maximum number of bytes cannot exceed the maximum transmission unit of the hardware device, otherwise, the hardware package will be reorganized. The increase step is 100 bytes. 32 packets are tested each time. If the network is busy, 128 packets are usually used for statistics. Tests show that the minimum round-trip latency increases linearly as the package size increases linearly.

Figure 1

The minimum round-trip latency trend uses the data for linear fitting and obtains the minimum round-trip latency trend fitting and line to obtain the current β I. Through this series of β values, we can use formula (2) to calculate the bandwidth capacity of each segment in the path.

According to formula (2), we tested the provincial cmnet network and the network of a city in Shanxi Province. According to the common network environment, we can obtain the test results in table 1. The results are compared with the statistical results of historical port traffic on the current network. It indicates that the variable packet length test method can test the actual network bandwidth capacity, however, because some network environments do not fully comply with the theoretical requirements, the 3rd test results should be smaller than the actual bandwidth capacity.

Serial number	Theoretical bandwidth	Test bandwidth	Network Environment Description
1	100 Mbit/s	41 Mbit/s	3 information points rack Transfer
2	100 Mbit/s	83 Mbit/s	Direct
3	100 Mbit/s	34 Mbit/s	Layer-2 switch exists
4	2.5 Gbit/s	1 536 Mbit/s	SDH
5	1 Gbit/s	319 Mbit/s	-

Table 1 bandwidth estimation test table

Serial number theoretical bandwidth test bandwidth network environment description

1100 Mbit/s41mbit/S3 information point rack Transfer

2100 Mbit/s83mbit/s direct

3100 Mbit/s34mbit/S has two-layer switch

42.5 Gbit/s1536mbit/ssdh

51 Gbit/s319mbit/s-

Even so, by analyzing the end-to-end network bandwidth estimation results, we can still find the network bottleneck and the quality of the network cable layout project, which makes sense for optimizing the data network.

1.2 defect of variable packet length Detection Method

The variable packet length detection method is based on the following prerequisites: first, assume that all the devices in the path work on the three-tier storage forwarding mechanism and support the TTL mechanism. Second, A single route is required. Third, the path is idle, and the minimum round-trip delay does not include the queue delay. We can see that the first precondition is the most demanding for the actual network, and the third hypothesis can be achieved through a large number of samples. After many tests, we found that the first condition was hard to meet mainly because there were some L2 devices with no IP addresses configured on the access network. These devices work on the L2 network, the TTL mechanism is not supported, so the round-trip time cannot remove the transmission latency caused by this hop.

In the current network test environment, we found that the network device's response speed to ICMP is also different. Some network devices use ICMP packets as low-level packages to respond to the operating system software design. Therefore, when the number of test packets is small and the network load is high, ICMP packets may have queue latency. This problem must be solved by increasing the number of test packages. However, due to the technical level and investment costs of data network management developers, microsecond-level timing has large defects, the test results are also affected.

2. Other common testing methods

There are other bandwidth estimation methods. Three methods are described below.

2.1 pptd (packetpair/traindispersion)

Pptd is a well-known method for measuring end-to-end bandwidth. The principle of the pptd method is to send a pair of data packets of the same size at a fixed interval at the source end, and then measure the time interval between the two packets at the receiving end, the end-to-end capacity can be calculated through this interval. This method requires no other interference traffic in the network, which is basically impossible in the actual network environment. Therefore, multiple measurements are required to filter out the interference Measurement packets. This method requires measurement at both ends at the same time. If it is used in the data network tube, the deployment cost is huge. It can be deployed in end-to-end deployment that is of great concern (for example, the end-to-end quality analysis of SP and SMS gateway, or test the connection bandwidth between the SP and the WAP Gateway.

2.2 Slops (self-loadingperiodicstreams)

Self-loadingperiodicstreams (slops) is another common bandwidth estimation method, which measures the end-to-end available bandwidth. The slops also sends data packets from the source end to the target end. The principle is that when the traffic rate of the test data exceeds the available bandwidth, the latency of the target end package will increase, when the test data traffic speed is similar to the available bandwidth, the latency will be relatively stable. This method is used to continuously send test data for traffic estimation. Therefore, it occupies a large amount of network resources and affects the normal business performance of the current network.

2.3 Topp (trainsofpacketpairs)

This method is similar to the slops method. It can not only measure the available end-to-end bandwidth, but also estimate the capacity bandwidth of the current path. However, this method also needs to send a large number of test packets, the network occupies a large amount of resources. In actual network tests, it is risky to pay a large cost.

3. Conclusion

This article describes how the data integrated network management system uses the ping call test to estimate the network bandwidth and provides the test results of the provincial network. The results show that this method is simple and feasible, and has positive significance for data network optimization. The article finally introduces some popular network bandwidth estimation methods, and analyzes the feasibility and possible use range. In the follow-up work, unidirectional latency analysis can greatly improve the accuracy of network bandwidth analysis, which is more meaningful for carriers in network promotion.