5.19. Neutron ML2/ODL Performance testing¶
- status
draft
- version
1.0
- Abstract
This document describes how control plane and network data plane performance testing is conducted on an OpenStack Cloud using the Neutron ML2/ODL plugin. The control plane performance is going to be analyzed in terms of response time and the parameters used to characterize the data plane performance are throughput in Mbps and latency in milliseconds and number of Request-Response Transactions.
5.19.1. Test Plan¶
Characterize the resource consumption, control plane response time and network performance of an OpenStack Cloud using the Neutron ML2/ODL Plugin. Control plane response time is measured when creating and listing common Neutron resources as well as during actions that require interaction between Neutron and other OpenStack services such as Nova. Network plane performance is measured for both TCP and UDP protocols, with throughput, latency and Request-Response benchmarks for TCP and Packets Per Second (PPS) and loss for UDP.
5.19.1.1. Test Environment¶
5.19.1.1.1. Preparation¶
TripleO is the vehicle used for deploying OpenStack with the ML2/ODL Neutron Plugin.
5.19.1.1.2. Environment description¶
The environment description includes hardware specs, software versions, tunings and configuration of the OpenStack Cloud under test.
5.19.1.1.2.1. Hardware¶
List details of hardware for each node type here.
Deployment node (Undercloud)
Parameter |
Value |
model |
Dell PowerEdge r630 |
CPU |
2xIntel(R) Xeon(R) E5-2683(28Cores/56Threads) |
Memory |
126 GB |
Disk |
2 x 1TB SATA |
Network |
1 x Intel X710 Quad Port 10G |
Controller
Parameter |
Value |
model |
Dell PowerEdge r630 |
CPU |
2xIntel(R) Xeon(R) E5-2683(28Cores/56Threads) |
Memory |
126 GB |
Disk |
2 x 1TB SATA |
Network |
1 x Intel X710 Quad Port 10G |
Compute
Parameter |
Value |
model |
Dell PowerEdge r630 |
CPU |
2xIntel(R) Xeon(R) E5-2683(28Cores/56Threads) |
Memory |
126 GB |
Disk |
2 x 1TB SATA |
Network |
1 x Intel X710 Quad Port 10G |
Additional Hardware for testing/monitoring/results
Performance Monitoring Host (Carbon/Graphite/Grafana)
Performance Results Host (ElasticSearch/Kibana)
5.19.1.1.2.2. Software¶
Parameter |
Value |
OS |
RHEL 7.3 |
OpenStack |
Newton |
Hypervisor |
KVM |
Neutron plugin |
ML2/OVS |
L2 segmentation |
VxLAN |
5.19.1.1.2.3. System Performance Monitoring¶
Record System performance metrics into a separate metrics collection/storage/analysis system. Suggested system would be a separate machine with Carbon, Graphite, and Grafana with dashboards for monitoring system resource utilization. To push metrics into the TSDB, collectd can/should be installed on all monitored machines. (Deployment, Controllers, and Computes)
5.19.1.2. Test Case 1¶
5.19.1.2.1. Description¶
Create and list Neutron resources such as networks, routers etc and measure the response time of the API.
5.19.1.2.2. List of performance metrics¶
Priority |
Value |
Measurement Units |
Description |
|---|---|---|---|
1 |
Reponse Time |
seconds |
Time taken for API to respond |
5.19.1.3. Test Case 2¶
5.19.1.3.1. Description¶
Create and list servers on a subnet and compare time taken for server to go into ACTIVE. This is an important test case as ML2/OVS and ML2/ODL have both take different approaches when notifying NOVA that the network is ready.
5.19.1.3.2. List of performance metrics¶
5.19.1.4. Test Case 3¶
5.19.1.4.1. Description¶
Using Browbeat to run Shaker, measure Latency of UDP small packets, TCP throughput and Request-responses in each of L2, L3-East-West and L3-North-South Scenarios.
5.19.1.4.2. List of performance metrics¶
5.19.1.5. Tools¶
Browbeat is used to orchestrate several Rally and Shaker scenarios and results are pushed to Elasticsearch for easy visualization through Kibana.
5.19.1.5.1. Setup¶
Deploy OpenStack Cloud using TripleO
Install testing and monitoring tooling
Gather metadata on Cloud
Run tests
5.19.1.5.2. Analysis¶
Review System performance metrics graphs during test duration to observe for stopping/failure conditions. Review testing harness output for test failure conditions. API response time and several statistics associated with it such as percentiles can be obtained from Rally and Shaker output. Compare ML2/OVS with ML2/ODL.
