Correct switch installation to avoid switch network faults

Many vswitch users have encountered many problems due to many reasons. Here we mainly analyze the vswitch network faults caused by incorrect Switch location installation. A vswitch is a key connection device in a LAN. It has a large number of vswitches in a LAN. In such a network environment, the administrator needs to reasonably deploy the vswitch based on its performance and characteristics, because your improper deployment will pose hidden risks in the later network O & M. I would like to share with you a case that causes a switch network failure due to improper installation location.

1. vswitch network faults

The computer room in a middle school encountered a very strange fault in a computer test simulation test: when students are doing a simulated test, client computers frequently experience network interruptions with servers. However, when a fault occurs, each client can PING the server with a latency of less than 10 ms, but the server is not visible from the network neighbors, in addition, when a fault occurs, each client computer can access each other through its network neighbors. Because the final simulated examination file of the student is stored on the server after being operated on the local machine, the network failure of the switch causes the simulated examination to fail.

2. Preliminary analysis, irrelevant to the server

After observation, it is found that the fault occurs when the number of students accessing the server increases. Therefore, the fault point is located on the server at the beginning, and the number of connections on the server may be set incorrectly. When the server is installed, the number of connections is set to 999, while the number of connections of servers working in other data centers is set to 256. Is the failure caused by excessive connections? So we reduced the number of connections to 512 and 256 in sequence, and the Failure remained after the server was restarted. Based on the experiment results, the failure assumption caused by the number of connections setting is excluded. The IBMX3600 server is newly purchased, so it is suspected that the fault is caused by a conflict between the new device and the installed examination system. So I immediately moved to another backup server, IBMX236, which can be used normally in another laboratory. However, when I switched to the backup server, the fault remained. So far, it is preliminarily determined that the fault has nothing to do with the server.

3. Locate the fault source. The problem lies in a switch.

So where exactly is the problem? In this lan, there are only three types of devices except Network cables: computers, servers, and switches. Based on tests that have been performed, the client computer and server faults have been ruled out. If the PING command is successful, the network cable is correct. Therefore, consider whether the fault is caused by a switch. Two types of switches are in use in the lab, namely, the netshitong DCS 2026 and H3C S1024R Switches of Digital China. In order to determine the fault point, 20 client computers and IBMX3600 servers in the lab are connected to one switch. First, the H3C S1024R switch is tested. When 20 clients are connected to the server by Long PING, the system also saves the questions to the server. The fault does not occur. Then, the 20 computers were connected to the DCS 2026 switch. When 15th client computers were stored on the server, the fault was reproduced, and the cause of the fault was initially determined on the switch.

4. Data Testing and in-depth analysis

Why does a switch of DCS 2026 cause a network fault? We decided to conduct a data test. In order to make the test result more appealing, we found a 3rd switch from CISCO, a 355 manufacturer, to perform a test in the same environment. During the test, the connection environment remains unchanged. All three types of switches use the factory default settings. A file with a size of MB is prepared on each client computer, after 20 computers are connected to shenzhou.com DCS 2026, H3C S1024R, and CICCO3550 in turn, the 845MB files on each computer are copied to the shared folder on IBMX3600, which is displayed by the Network neighbors. The EtherPeek packet capture software is used on the server to analyze all data flowing through the ports connected to the server.

(1). Test the data of the DCS 2026 Switch

When the client computer files are copied to the server in sequence by DCS 2026, only l2 computers can be copied to the server at the same time. When 13th computers copy files to the server, the network connection is invisible. In this case, the client's long Ping SERVER is normal. At the same time, the port traffic on the server also decreases from 62.284 Mbits/s at 12 times to 41.183 dbits/s. By analyzing the captured data packets after a vswitch network failure occurs, it is found that the traffic data of 41.183 dbits/s is almost all small data packets such as Ping packets.

This phenomenon indicates that when the traffic on the vswitch DCS2026 reaches about 63 Mbits/s, the large-size data packets are discarded and the Small-size data packets such as Ping packets are forwarded normally, this is why clients can Ping the server when a fault occurs, but cannot access the server through network neighbors. In order to eliminate a single port failure, the test results are the same for the test of changing the port on cyi DCS 2026.

(2). Test the H3C S1024R switch data

In the same hardware and software environment, when the H3C S1024R switch sends data to the server on 17th computers connected to it, a fault occurs. At this time, the traffic is reduced from about 73 Mbits/s to 42.23 Mbits/s, which is different from that of Digital China. The Ping SERVER on all computers is normal, and the nine client computers that transmit files to the server work normally. The connection between the last eight computers and the server's network neighbors is interrupted. The test results show that the H3C S1024R switch does not discard all large-size data packets, but the switch processes the forwarded data in priority.

(3). Test the data of the cisco3550 switch.

When the CISCO3550 switch is tested, the 2O computers connected to it can send normal files to the server while the server is pinged for a long time, the port traffic reaches 101.4 Mbits/s. The above tests are performed when each type of switch is connected to 20 client computers. Then, we connect only the server and a client computer on each switch in turn. the client sends the same 845MB file used in the previous test to the server. The time consumed by CISCO is 180 s, that is, the time consumed by netc dcs is 2026 s, and that of the H3C S1024R switch is 179 s. The test results show that, when the switch load is small, the traffic of CISC03550 is similar to that of the smart China Smart Access Gateway (DCS) Port 2026, while that of the H3C S1024 is slightly weaker.

Finally, we tested the data traffic during data storage operations. Through packet capture analysis, we performed data storage operations on a single computer (saving files to the server ), within 25 s, we continued to click the Save button for open project files to save the files to the server. The average traffic reached 496 kbits/s. There are a total of 7 vswitch-level connections in the IDC. each vswitch has 24 ports. Based on the test data, the total traffic for concurrent operations on the port connecting to the server during the examination is calculated as follows: 496x7x23 = 76.57 Mbits/s. The traffic exceeds the maximum traffic of 2026 Mbits/s on a single port of the Shenzhou digital 62.284 switch. Through the above test results, we can conclude that the failure of the switch is caused by the traffic limit on the port of the smart China 2026 switch. The root cause is that the vswitch discards large-size data packets when the data volume to be forwarded is large.

5. Solution

A network fault solution for a vswitch is as follows: add a cisco3550 or similar aggregation switch to the device selection. you can also add a gigabit module to an existing vswitch and connect the server to it. The network topology is changed from a hierarchical connection to a star architecture. The access switch can continue to adopt H3C S1024R or a vswitch with similar performance. Connect all access switches to the aggregation switch.

Conclusion: It is very important to fully understand the network traffic mode and the actual maximum traffic on the port of the switch used. The purpose of using a vswitch is to minimize and filter data traffic in the network. Therefore, if a vswitch in the network needs to forward almost all received packets due to its improper installation location, instead of optimizing the network performance, a vswitch reduces the data transmission speed and increases the network latency, even because the manufacturer considers the design, discard a specific type of data packet, which leads to various strange faults in network usage.