2016 OpenStack Barcelona – Maheshkumar Pandurangan – On Demand scale and performance Data Analytics
Half Information is no information. There is a good amount of control plane scale and performance metrics collected by Rally and utilization metrics collected by tools like Monasca/collectl in the context of scale and performance test runs.
Unless these data are correlated there is no meaningful information. In this presentation we propose to aggregate the data from disparate data sources to a centralized analytic platform to perform on demand correlative analysis like
– what is the memory utilization trend of neutron L3 agent, in a scale of ‘n’ ports using CVR vs DVR
– What is the I/O utilization on the compute node for VM boot during scale, with windows image vs linux image
– Compare cinder-volume performance against different backends for a scale run of ‘n’ volumes
– How is the Cloud Controller CPU utilization and nova boot throughtput during a scale run of ‘n’ VMs against different processor clock speed
by vBrownBag
openstack
Good thought …