-
New Feature
-
Resolution: Fixed
-
Major
-
None
-
None
Implement functionality that will allow for the measurement of the performance impact of individual features for a given vSwitch. The scope of this work does not cover implementing support for features not currently supported in VSPERF eg. conntrack.
Sample usage:
The user specifies a topology, and a list of test parameters. Eg.:
PVP - vHost MQ, 1k flows, VxLAN tunneling, IMIX packets,2C.
This will trigger 6 (N features + 1) VSPERF test runs, result for each gathered:
- Baseline (default PVP config)
- + MQ (default was 1 queue)
- + 1k flows (default was 1 flow)
- + VxLAN tunnel
- + IMIX packets (default was 64B)
- + 2 cores (default was 1 core)
The above assumes that cumulative performance will be measured ie. once we enable a feature it isn't turned off for the next test run. This could be changed/specified when triggering the test run eg. cumulative=true/false
For the =false case, we enable one new feature per test run and disable all others.
Results are collated and worked on to produce a summary which reflects the impact of individual features like so:
Feature | Perf | Individual impact | Total impact |
Baseline | 7Mpps | 0% | 0% |
MQ | 8.5Mpps | +21% | +21% |
1k flows | 7.5Mpps | -12% | +7% |
VxLAN | 4Mpps | -47% | -42% |
IMIX | 4.5Mpps | +13% | -36% |
2C | 8Mpps | +78% | +14% |
Sample command line could be:
./vsperf --matrix topology='pvp',params='x,y,z',cumulative='true'