DRAFT

Notes on benchmarking (with Rally) the performance of creating and querying samples against the mysql datastore.

Overview

The general idea is:

• to install two different devstacks (one stable/icehouse the other master (and thus Juno))
• detune each devstack to run the minimum required set of services
• configure ceilometer-api to run under mod_wsgi (to allow concurrent requests)
• run a RallySampleScenario against the devstack

Caveats

• Rally's use of the database can flake out when dealing with large result sets.
• Setting up Icehouse is a good deal more complex than Juno
• Volume of data is still too low for realistic measurements.
• Do Juno first as it puts some useful artifacts it place.

Make Sure

• The database is dropped between test runs.
• ceilometer-api port is not blocked by apache when starting up icehouse

Juno/master

• Clone devstack, switch to branch ... <- fill in
• Make a reasonable JunoLocalConf (which includes rally and sets ceilometer-api to use mod_wsgi)
• ./stack.sh
• Kill extraneous services
• Confirm rally: rally deployment check
• Confirm ceilometer: ceilomter sample-list -m image
• Use the sample scenario rally task start ./scenario.json
• Record results

Icehouse

• Clone the stable/icehouse branch of devstack.
• Make a reasonable IcehouseLocalConf
• Make sure there's no localrc (which would override local.conf)
• ./stack.sh
• Configure Rally for existing installation
• Kill extraneous services (including ceilometer-api, but leaving keystone)
• Re-enable ceilometer under mod_wsgi (details to follow)
• Restart apache.
• Confirm samples with ceilomter sample-list -m image or similar
• Use the sample scenario rally task start ./scenario.json
• Record results