-
Stéphane Diemer authoredStéphane Diemer authored
Benchmarking
This documentation explains how to benchmark a MediaServer to get the maximum viewers that can be handled. The benchmarking tool is based on Locust.
Prerequisite
- Envsetup and ansible Help to get envsetup.
- An SSH access to the machines that will be used to make requests on MediaServer (you can use the workers for that).
- A video on demand or a live stream to test. To create a live stream for the test, please read the section
Prepare a live page
of this documentation.
Note that the machines used by the benchmarker must have an access to UbiCast packages (to get the ubicast-benchmark
package).
Hardware requirements
To be able to test a server, the benchmarking tool must use systems with enough CPU, RAM and bandwidth.
You can calculate the required CPU and RAM by extrapolating the following data:
- A 20 threads CPU @ 3.0GHz is required for the benchmark worker to test a hatch rate up to 50.
- A 20 threads CPU @ 3.0GHz is required for the benchmark worker to test up to 3000 viewers.
- 6 GB of RAM is required for the benchmark worker to test up to 3000 viewers.
- A maximum of 5000 viewers can be tested on a single for the benchmark worker system (due to ephemeral ports limit).
- The benchmark server will use approximately 25% of the CPU allocated to workers (for example, if the benchmark worker has 4 CPU, the benchmark server must have 1 CPU at the same speed).
Prepare a live page
If you want to benchmark the video on demand playback, ignore this section.
This section explains how to get a live page to get an oid
for the bench.
This live page can be used for the bench even if no video stream is sent to the server when the bench_dl_streams
is set to false
.
Steps:
- Login in MediaServer
- Click on the
Add content
button - Click on
advanced
- Click on
Add a live stream
- Click on
For UbiCast or other recorders
- Set a title, for example
live bench
- Click on
Add live stream
- Click on
Edit
- Check
published
- Set
live status
toongoing
- Click on
Save changes
- Open the
Permissions
tab - Set the access to
yes
forNon authenticated users
- Click on
Save changes
- Open the
Resources
tab - Click on
Get encoder settings
- Use the
oid
value in the page url (for examplel125f58fbb8c655nth76
) in yourbench_oid
setting.
Inventory
Create a new inventory with following configuration.
Hosts file
This file should be located in <your inventory>/hosts
.
[bench_server]
worker1.test.com
[bench_worker]
worker2.test.com
The system targetted with bench_server
will host the Locust server. This system must be unique.
The systems targetted with bench_worker
will host the Locust workers. Multiple workers can be used.
Group vars
This file should be located in <your inventory>/group_vars/all.yml
.
bench_server: <The benchmark server IP or host name>
bench_host: <MediaServer URL. Example: "msauto.ubicast.net">
bench_oid: <Media OID. Example: "l125f58fbb8c655nth76">
bench_user: <MediaServer account username. Can be empty if the media access is not protected. Example: "test">
bench_password: <MediaServer account password. Can be empty if the media access is not protected. Example: "pwd">
bench_host_api_key: <MediaServer master API key>
bench_dl_streams: <Download video streams or not ("true" or "false", "false" is default)>
Install the benchmarker
# ansible-playbook -i inventories/<your inventory> playbooks/bench.yml
This playbook will install everything you need: benchmark server and benchmark workers.
Start the benchmarker
Disable anti-ddos in Nginx configuration if any (usually in /etc/nginx/conf.d/limits.conf
).
If the server hosting the bench server is using a firewall, disable it to be able to access the locust interface. For example, if ferm
is installed: systemctl ferm stop
.
If you want to test with a video stream, you can start it with using the docker container:
cd /usr/share/ms-testing-suite && make run_live
Or directly by hand:
/usr/share/ms-testing-suite/ms_live_streamer.py /etc/mediaserver/bench-streaming.conf
Go with your browser on http://<bench_server>:8089
. You might have to use an SSH tunnel to access this port.
Set the number of viewers and the hatch rate (usually the hatch rate is 0.25% of the viewers count to reach 30% of the total count in 2 minutes) and start the bench.
Watch for /var/log/error.log
and /var/log/access.log
(warning: this file is buffered by default, you can disable buffer in nginx.conf
).
Restart Locust server & workers
If you need to restart both server & workers, you can launch the playbook with the tag prepare-bench
:
# ansible-playbook -i inventories/<your inventory> playbooks/bench.yml -t prepare-bench
/!\ EXPERIMENTAL : Elasticsearch + Kibana + metricbeat
Kibana + metricbeat allows you to monitor and display statistics about the infrastructure. This feature is experimental.
First you need to add a new group to your hosts files:
[elastic]
elastic1.test.com
Then launch the playbook bench-monitoring
which will install both the elastic suite on the elastic host, and metricbeat on mediaserver & postgres servers.
# ansible-playbook -i inventories/<your inventory> playbooks/bench-monitoring.yml
Your kibana instance will be accessible at http://elastic1.test.com:5601
.