Overview of server testing methods from the perspective of performance and functionality

This article describes the server testing method from two parts: performance testing and functional testing. Each test is performed three times in the process, and the server and tester are restarted between each test. The results are averaged three times to ensure the accuracy of this server testing method.

The test tool uses Avalanche2200 provided by Spirent Communications. Avalanche 2200 simulates actual users sending requests to the server (including SMTP and POP3), and gives specific and detailed test results based on the responses. It has the following features: it can simulate hundreds of clients sending requests to the server; it can simulate real network application situations; it can generate 20,000 connections/second, which is enough to meet the needs of the test.

The monitoring end is installed with Windows 2000 Server with SP2. The test environment is shown in the figure below. Two servers under test (with the same hardware and software configuration) are directly connected and connected to the test port of Avalanche 2200 through switches (Harbour's Ammer24 switches). For easy management, we use Avocent KVM switches, which can easily control multiple servers remotely.

In order to simulate the test environment and test model to the actual operation of the mail system as much as possible, we have made full considerations in the test method. First, the mail system is established by two servers. If one server only has local delivery but no remote delivery, the pressure on the mail system is not great, and the actual user usage will definitely include remote delivery, so we conducted a mail system test on two servers. Second, multi-domain testing is adopted. The two servers are configured with independent DNS servers and multiple domain names (5 domain names) are specified. Third, multi-account testing is adopted. We set up 200 accounts in each domain, a total of 1000 accounts. During the test, users of the two servers sent mail to each other, accounting for 90% of the total, and at the same time, users within each server sent mail to each other, accounting for 10% of the total. Fourth, the size of the mail sent meets a certain distribution. The sizes of the mails sent are: 400, 1000, 2000, 4000, 10000, 20000, 40000 and 100000 bytes.

The specific test process consists of four stages. The first stage is the warm-up stage, in which the number of mail requests sent by Avalanche gradually increases from 0 to 10/s; the second stage is the gradual pressure stage, in which the number of requests gradually increases to 510/s (this value exceeds the maximum capacity of all participating mail systems); the third stage is the maintenance stage, in which the number of requests remains unchanged; the fourth stage is the decline stage, in which the number of requests decreases from the maximum value to 0. Finally, the maximum number of concurrent requests of the server is obtained, and the success percentage, round-trip delay and other results are also provided.

The various operations of the POP3 protocol include CHECK, LIST, RETR, DELETE, etc. The test methods for each are roughly the same as SMTP. The mail systems tested this time all provide account security protection functions. For the same email account, only one user is allowed to receive mails using POP3 at the same time. To perform stress testing, a large number of POP3 connections must be initiated at the same time. Therefore, we only verified DELETE (each account only logs in once to perform DELETE operations).

After receiving the email, the server only stores it locally and creates an index before sending a "success" message to the user. Therefore, it does not reflect when the user actually receives the email. Therefore, we designed a method to investigate how long it takes for an account to receive emails. We wrote a program ourselves, using 100 accounts on a server (20 accounts for each domain) to send 1,000 emails to 100 accounts on another server. Each email is 20K bytes long. After the sending is completed, the account to which the email was sent is checked immediately to obtain the number of emails received, and then checked again after 5 minutes. If it is not completely received, then check again after 5 minutes until all 1,000 emails are received.

Each test was performed three times, with the server and tester restarted between each pass, and the results were averaged over the three runs.

In terms of functional testing in the server detection method, we focus on security, multi-domain management, and mail storage, and also examine the system operating platform, system support capabilities, user service capabilities, mailbox management capabilities, etc. We also examine the system management, client management, and the ease of use of the entire system.

The above are the detailed steps of this server testing method. I hope readers can learn and apply it flexibly and it will be of practical help to everyone’s work and life.