Shulou(Shulou.com)06/03 Report--

Previously, I have written a RFC2544 performance test content, which is followed by the RFC2544 performance test item throughput to further explain how the network layer throughput performance should be tested.

The throughput performance data will eventually be fed back to the target groups: themselves, developers, product personnel, and users. First of all, we have to consider how all kinds of people measure the performance of the product from which angle. Here we give two testing methods both internally and externally.

The following is the test method chosen by the users encountered by the author in the external test project: there must be some shortcomings, welcome to add:

A, regular RFC 2544 test, udp bi-directional symmetric packet test of 64,128,256,512, 1024, 1280, 1518 and other seven kinds of byte packet length

B. Regular RFC 2544 test, udp one-way packet test with seven byte packet lengths of 64,128,256,512, 1024, 1280, 1518, etc.

Note: the test methods of an and b are very popular and recognized over a period of time. For example, qualification testing, finalist testing and project testing all use the above standards as the network layer testing standards, but with the passage of time, the testing methods become familiar and popular, and people think of more testing methods, such as:

C. Regular RFC 2544 test, udp bidirectional packet test of 64,67,128,256,512,1024, 1280, 1517, 1518 byte packet length

D, regular RFC 2544 test, udp one-way packet test of 64,67,128,256,512,1024, 1280, 1517 and 1518 byte packet length

Mixed test with a certain proportion of e, 64,512 and 1518

Ip message testing of f, 64,128,256,512,1024, 1280, 1518

For the above customer testing methods, we need to clarify the support of our current products. Take the firewall as a simple example, different manufacturers have different ways to achieve their own forwarding.

1. Traditional packet filtering firewall: all packets need to be filtered without choice, so there is no problem with the above methods of forwarding, and it has little impact on the performance.

2. Stateless connection firewall: the above f test method may cause forwarding and performance problems. This kind of firewall wants to establish a connection, the premise of connection establishment is five tuples, but at this time the test message is ip message, how should the connection be established? Is it necessary to perform route, nat, acl and other logic every time in the process of forwarding, and will it affect the performance?

3. Stateful connection firewall: fast forwarding mode and default forwarding mode will be made for performance improvement through stateful connection firewall. For these two modes, if most manufacturers want to enter fast mode, they all need two-way packets on the connection, so when these two conditions are limited, will the test methods b, d, f show abnormality? Add a little extra: after entering the connection in fast forwarding mode, there is a sharding packet. Is there a problem with forwarding?

4. Next-generation firewall: as we all know, the premise of next-generation firewall is based on application identification, high performance, cloud feature analysis and so on. What problems will the above testing methods bring? Let's analyze it, everyone.

Next is the internal test: under the premise of the above test methods a, b, c, d, f, g, we usually add different operating mode tests, such as: layer 2 (vlan, bridge, bond, bypass, etc.), layer 3 (routing, bond, subinterface, etc.), respectively verify the performance of fast forwarding mode, default forwarding mode, and performance differences under different working modes, and make clear the rationality of the differences. Based on the above tests, the functions such as routing, nat and policy are extended to test the impact of various functions on forwarding. In addition, in the past experience, there is another parameter that has a great influence on throughput performance: the number of concurrent connections. Testing the performance of throughput under different numbers of concurrent connections is likely to find performance problems caused by the number of concurrent connections. For the above tests, please record the detailed test process data for easy viewing and future tracking and recall.

The record of the test point is as follows: (only 2 items are simply listed)

Second floor:

Vlan+ Express forwarding Mod

1. Test results: such as package forwarding rate and delay

2. Testing process: for example, cpu utilization, memory utilization, concurrent connections, perf records and other data conducive to performance analysis

Vlan+ Express forwarding mode + features (routing, nat, policy, etc.)

Vlan+ default forwarding mode

Vlan+ default forwarding mode + function (routing, nat, policy, etc.)

Three floors:

Routing + fast forwarding mode

1. Test results: such as package forwarding rate and delay

2. Testing process: for example, cpu utilization, memory utilization, concurrent connections, perf records and other data conducive to performance analysis

Routing + Fast forwarding mode + features (routing, nat, policy, etc.)

Route + default forwarding mode

Routing + default forwarding mode + features (routing, nat, policy, etc.)

Combined with the previous article, check the forwarding rates of fast forwarding mode and default forwarding mode in different byte package, and compare various models and platforms to see if there is a certain relationship.

In addition, please note that the tested data must reflect on why it is such a data and how it comes from, so that it can be improved.

Internal testing lists a lot of data that need to be tested, just to reflect the performance of the product more completely, but with so much performance data and so much workload under the tight testing resources, please think about how to do it.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.