DragonFlow Data Plane Performance Testing Spec

DragonFlow Data Plane Performance Testing Spec

This spec describes the data plane performance testing of DragonFLow.

The spec includes testing scenarios, network link quality definition, testing methodology and testing tools and its usage.

Testing scenarios

DragonFlow data plane performance benchmark in different scenarios:

  • East-West data plane performance testing
  • North-South data plane performance testing
  • Source-NAT (SNAT) data plane performance testing
  • High load network traffic
  • High go-to-control traffic
  • Monday morning storm
  • And more..

East-West data plane performance testing

In this test we want to measure the network link quality in east-west traffic in different scenarios, both in single node and multi-node, combined with high number of flows handled by DragonFlow.

Environment setup

  1. Create two new private networks
  2. Create 100 vms/dockers on each node
  3. Create a vm with iPerf client
  4. Create 4 vms with iPerf server:
  1. Same host traffic
  • L2 scenario
  • L3 scenario
  1. Cross host traffic
  • L2 scenario
  • L3 scenario

Tests description

Generate 10 consequence tests with 5 seconds sleep between each test towards each iPerf server. Run this test with different network encapsulation protocols:

  • STT, GRE, VxLan, Geneve
  • Run this test using Vlan tagging

Tests types

Measure line Bandwidth, Datagram loss, Jitter and Latency.

Total number of tests

4 x 5 x 4 = 80 tests

North-South data plane performance testing

In this test we want to measure the North-South network link quality, combined with high number of flows handled by DragonFlow controller.

Environment setup

  1. Create two new private networks
  2. Create 100 vms/dockers on each node
  3. Create a vm with iPerf client
  4. Create 2 vms with iPerf server with a floating IP:
  • Same host as the iPerf client.
  • Cross host as the iPerf client.

Test description & scenarios

  • Generate 10 consequence tests with 5 seconds sleep between each test towards the server.

Tests types

Measure Bandwidth and rate, Datagram loss, Jitter and Latency

Total number of tests

2 x 4 = 8 tests

Source-NAT (SNAT) data plane performance testing

In this test we want to measure the SNAT (source network address translation) network link quality, combined with high number of flows handled by DF.

Environment setup

  1. Create two new private networks
  2. Create 100 vms/dockers on each node
  3. Create 2 vms with iPerf client.
  4. Run iPerf server on external IP

Test description & scenarios

  • From each of the iPerf clients, generate 10 consequence tests with 5 seconds sleep between each test towards the server.

Tests types

Measure Bandwidth and rate, Datagram loss, Jitter and Latency

Total number of tests

2 x 4 = 8 tests

Testing methodology

  • For measuring DragonFlow networking improvement/overhead, all tests have to be executed with & without DF (with DVR. OVN as well?).
  • The created VMs should report when there are up, so it will be possible to count the successfully created VMs in the automation.

Tools & usage

We will use iPerf for most of the tests. iPerf is a tool to measure the bandwidth and the quality of a network link.

Also we will use scripts that we will develop in order to automate the environment setup tests execution.

iPerf

The network link is delimited by two hosts running iPerf.

Bandwidth performance

iperf -c Dest_IP

Reverse mode bandwidth performance

iperf -c Dest_IP -r

Datagram loss & Jitter

Client: iperf -c Dest_IP -u

Server: iperf -s -u -i 1

Latency (Ping)

ping -c 10 -i 0.2 -w 3 DEST_IP

Creative Commons Attribution 3.0 License

Except where otherwise noted, this document is licensed under Creative Commons Attribution 3.0 License. See all OpenStack Legal Documents.