40GbE upgraded network performance test

Higher processor performance and increased network traffic brought about by virtualization. The best way to handle this traffic is to upgrade the 10GbE data center link to 40GbE. However, faster backbone links do not necessarily provide high performance. Upgrading to 40GbE requires network performance tests involving multiple factors.

Performance Considerations before 40GbE upgrade

After IEEE announces the 802.3ba standard, that is, the 40/100G Ethernet standard, network providers have launched a series of products designed to improve the line rate. They want to solve the bottleneck between racks in the 10GbE link. This bottleneck problem is that engineers use the 10GbE interface to replace the 1GbE server network interface, and 10GbE traffic flow from each server to the top of the rack.

There are many other problems and Influencing Factors to consider before upgrading the Backbone link:

Backplane:In some cases, although the supplier has introduced a higher-capacity interface card, the throughput of the vswitch and vro backplane may not support the full capacity operation of all the upgraded interfaces. Therefore, before replacing the interface card, you should carefully check the product specifications and consult the independent product test results.

The good news is that although the data transmission rate is improved, the higher protocol layer is not affected, so testing in this new environment is simplified. All these familiar exchange and routing technologies continue to work and do not need to be reconfigured in large quantities.

Upgrading to a higher speed backbone may expose other network problems. After the new hardware is installed, it is best to perform a series of network tests to determine whether the maximum performance improvement has been achieved.

Upgrade to the 10GbE server interface and 40GbE backbone at the same time:It is best to upgrade the backbone first, and then upgrade the server interface after confirming that the backbone is successfully running. Finally, after upgrading the server interface to achieve higher server throughput, check that the server vswitch does not become a bottleneck.

Upgrade the network performance test device:Many existing testing and monitoring products do not fully process higher data rates. There are some products on the market that generate load and monitoring performance for 40/100 GbE environments, such as products from test equipment vendors Spirent Communications and Ixia.

After Upgrade:What is network performance testing required?

End-to-End throughput test:This test emphasizes the main reason for the upgrade: a large amount of data needs to be transferred from the server to the server.

In the test, we should consider passing the traffic through a single backbone switch and through multiple switches. Traffic from multiple servers must be generated using a hardware load simulator or simulated by software running inside the virtual machine.

Network Performance Testing should include the situation where the Server Load balancer sends extremely short packet streams, full-length Packet streams, and burst packet streams. Network Monitoring hardware and software are used to identify the location of Long Packet queues or packet loss caused by bottlenecks.

At the same time, it is necessary to test sensitive applications to ensure that the increased network throughput does not cause problems. For example, the cluster heartbeat packet may lose order after it passes through the upgraded rack and the backbone switch.

TCP uses the congestion control algorithm to control and limit the rate of transmitted packets. Therefore, it may be necessary to modify parameters to achieve the full throughput supported by the network, such as the size of the sending and receiving windows.

However, because each network and usage mode are different, it is impossible to specify the required tests for a specific network. Since it is impossible to test each possible combination of data streams in the business network or the path adopted by each traffic, it is best to establish a more general test with a large amount of data packet traffic between racks.

Test jitter:Changing the packet arrival rate may cause an unacceptable pause in voice and video transmission. When other simulated applications generate loads on the network, it is best to use the hardware test generator to create an equal distance data packet stream. Measure the change in the arrival time to determine whether the jitter remains acceptable. If too much jitter is detected, it is necessary to modify the VLAN priority configuration for packets that require continuous delivery rate.

Fiber Channel Ethernet (FCoE ):This type of aggregation requires zero packet loss. Disk traffic increases with the increase in application performance and network load. Determine that the priority setting of the low-bandwidth network hardware is still suitable for the upgraded network.

The cable test should not be a 40GbE upgrade problem:The cable test is not a problem during 40GbE upgrade because the cable used for the 10GbE link must be replaced. Early Ethernet upgrades require testing existing cables to ensure they support high speed.

The cable supporting 40GbE consists of 10 fiber bundles because the supplier cannot transmit 40GbE through a single fiber bundle. Therefore, traffic in each direction is divided into four units, each carrying 10GbE. Two cables are not used, so the replacement must change with the change of the switch and router interface. Integrity of new cables should be tested during installation.

The main purpose of the network performance test is to maximize the benefits of upgrading the network link: The test can help assess when the next upgrade is required. There is no doubt that the upgrade to 100GbE will be in the near future.