d73fe40f4f
https://thirdparty-logs.rdoproject.org/jenkins-browbeat-quickstart-ocata-baremetal-no_vlan-46/undercloud/home/stack/browbeat/results/20170518-220119/all-rally-run-0.html.gz So if you look at these test results you can see that the 12 concurrency doesn't even run because we knock over the endpoint. We only have two quad core computes for this and they always get knocked over. Another problem is that Browbeat only checks if the rally task generates a file, so if the endpoint is down this isn't described as a test failure, but if the endpoint returns an error (arguably better than being down) Browbeat will fail the test. Overall I don't like the randomness of this. Change-Id: I27f80c8f9acc77c74a40a715e28af54ec3354456
Shaker Data Plane Performance Dashboards
Two dashboards have been provided with Browbeat for Shaker.
Browbeat Shaker Scenarios with Throughput vs Concurrency -------------------------------------------------------This Shaker dashboard aims to present data plane performance of OpenStack VMs connected in different network topologies in a summarized form. Three distinct visulizations representing L2, L3 E-W and L3 N-S topologies along with the corrensponding markdown to exaplain each visualization make the "Browbeat Shaker Scenarios with Throughput vs Concurrency" dashboard. For each network topology, average throughput for TCP download and upload in Mbps is expressed vs the VM conccurency (number of pairs of VMs firing traffic at any given moment). For example, in the L2 scenario if the average throughput is 4000 Mbps at a concurrency of 2, it means that each pair of VMs involved average at 4000 Mbps for the duration of the test, bringing the total throughput to 8000 Mbps(avg throughput*concurrency).
Browbeat Shaker Cloud Performance Comparison
This Shaker dashboard lets you compare network performance results
from various clouds. This dashboard is ideal if you want to compare data
plane performance with different neutron configurations in different
clouds. For each topology, a visualization comparing
tcp_download and tcp_upload per cloud name and
a visualization comparing ping latency per cloud name is generated in
the dashboard along with instructions in markdown for advanced filtering
and querying.
Note
You can filter based on browbeat_uuid and
shaker_uuid to view results from a specific run or shaker
scenario only and record.concurrency and
record.accommodation to filter based on the subest of the
test results you want to view.