OpenStack Mitaka Design Summit

Ceilometer has got significant performance, scalability and reliability problems.

Is there a magic formula to fix Ceilometer?

We added Monasca as a storage layer to Ceilometer and … we got Ceilosca!

The preliminary performance results were really encouraging; first, we could store in excess of6 million samples with a single node deployment and second, we recorded reasonable query response times:

Query (6M samples)                        Average Time (s)

ceilometer sample-list -m instance      2.722209
ceilometer meter-list                            0.316250
ceilometer resource-list                     28.852132
ceilometer statistics -m image            3.066389

How we did it. 
Ceilosca leverages the Ceilometer data collection pipeline via a Ceilometer-Monasca Publisher and the Ceilometer API via a Ceilometer-Monasca Storage Driver. This approach addresses the problems with Ceilometer, while still supporting a fully compatible Ceilometer v2 API and leveraging the excellent work in the Ceilometer OpenStack data collection pipeline, which is not addresed in Monasca.


What we will present:
We are working toward the implementation of Ceilosca and we will make available all the code as part of the Monasca code base in Stackforge. We are also planning to demo the solution and provide some performance evaluations of Ceilometer vs. Ceilosca (*) and illustrate the penalties introduced by the cross services overhead and format conversions.