← Back to team overview

openstack team mailing list archive

Re: [ceilometer] requesting feedback on priorities for metering data collection


Thank you very much for the answers, they are of great help.

I will plan on attending Thursday meeting so I can see the Essex discussion.

Glaucimar Aguiar

From: Doug Hellmann [mailto:doug.hellmann@xxxxxxxxxxxxx]
Sent: quarta-feira, 25 de julho de 2012 14:29
To: Aguiar, Glaucimar (Brazil R&D-ECL); openstack@xxxxxxxxxxxxxxxxxxx
Subject: Re: [Openstack] [ceilometer] requesting feedback on priorities for metering data collection

On Wed, Jul 25, 2012 at 10:45 AM, Aguiar, Glaucimar (Brazil R&D-ECL) <glaucimar.aguiar@xxxxxx<mailto:glaucimar.aguiar@xxxxxx>> wrote:

Thanks for sending this out. I am interested in this project and as I could not have it working yet with my Essex release of OpenStack

Some of the pieces work with Essex, but the actual daemons don't, yet. We are planning to discuss this at our meeting this Thursday (http://wiki.openstack.org/Meetings/MeteringAgenda).

I would like to better understand the metrics collected you mention below.
I have being reading the documentation and code extensively and would like to count on your help to better understand some aspects of the ceilometer.

Some questions below, sorry if they sound so easy for you but if you could help me with these, this would be of a great help for me.

These are good questions, thanks for bringing them up!

When you say CPU usage meters, does it mean ceilometer is already capable of reporting the cpu consumption over an hour?

We are currently polling libvirt periodically to get the total CPU time for an instance. The frequency of polling is adjustable, but I think the default is 10 minutes. The value returned by libvirt represents the total time for the instance, but a client could pull the data and perform the calculation over a period of time. The API we have defined allows for querying within a time range, for example.

Does this take into account the number of CPUs of the instance or this should be derived from the instance data?

The number of CPUs in an instance is part of the metadata associated with the counter, but does not enter into the calculation of CPU-time as far as I know. We ask libvirt for that info, and just use what we're given.

Thank you very much for any help in increasing my understanding of the behavior of this component. I can also contact you in IRC if you prefer.

I hope my answers help,

Glaucimar Aguiar

From: openstack-bounces+glaucimar.aguiar=hp.com@xxxxxxxxxxxxxxxxxxx<mailto:hp.com@xxxxxxxxxxxxxxxxxxx> [mailto:openstack-bounces+glaucimar.aguiar<mailto:openstack-bounces%2Bglaucimar.aguiar>=hp.com@xxxxxxxxxxxxxxxxxxx<mailto:hp.com@xxxxxxxxxxxxxxxxxxx>] On Behalf Of Doug Hellmann
Sent: quarta-feira, 25 de julho de 2012 11:24
To: openstack-operators@xxxxxxxxxxxxxxxxxxx<mailto:openstack-operators@xxxxxxxxxxxxxxxxxxx>; openstack@xxxxxxxxxxxxxxxxxxx<mailto:openstack@xxxxxxxxxxxxxxxxxxx>
Subject: [Openstack] [ceilometer] requesting feedback on priorities for metering data collection

When the ceilometer project started after the Folsom summit, we compiled a list of metrics to be collected [1]. We have reached a stage in the project where we are ready to start implementing more of these meters, and would like some input from the rest of the community about priorities. We have implemented the instance counter, disk I/O, and CPU usage meters, and have begun working on collecting some of the network statistics. If you are interested in using ceilometer for tracking usage in your cloud deployment, please let us know which other metrics you will find especially useful.


1 - http://wiki.openstack.org/EfficientMetering#Meters