Skip to content

Improve "Deployment using manifest" and "Kepler Model Server" documentation #168

@juangascon

Description

@juangascon

Hello.
I am using Kepler locally (kind) since at least a year ago and now I would like to use custom models that fit the VMs deployed by the hyperscalers Kubernetes flavors.

I would like you to give some more details in the "Deploy using manifest" section In that section, you give the OPTS to build the manifests but the descriptions are really short. I do not know exactly what does each option.

Also, there is no link with the Kepler Power Estimation Deployment chapter. It should be nice to put in this chapter what are the related options to build the manifests that correspond to each deployment.
I mean, if I would like to deploy the "Minimum deployment connecting to Kepler Model Server", what should I use as OPTS ?

Another nice-to-have would be some examples for the links to use to select a power model in the Select power model section. It is really hard to know what should be put there.
I would like more details on the use of offline and online models.

  • How can we reach the online models?
  • What we have to deploy and install to use the offline ones?

I am really struggling to arrive to deploy Kepler as it should with the good necessary components because I do not get what I have to do.
Maybe is right before my eyes but I do not get the picture.

I have some Data Scientists that want me to validate that we are using the good models and not the default ones.
In the documentation it is said that :

Kepler container image has preloaded acpi/AbsPower/BPFOnly/SGDRegressorTrainer_1.json model for node energy estimate and rapl/AbsPower/BPFOnly/SGDRegressorTrainer_1.json for Container absolute power estimate.

I would like to load another "Pre-trained Power Model" since, as it is said in its section of the Kepler Power Model, there are a lot of models in kepler model db; although I do not know what should I take among the available 4.
I am pretty sure that I lack some Kubernetes knowledge to find out by myself what I should do.
This is why I ask you for help with a little more doc or examples of ConfigMaps and links.

Thanks a lot for all the work.

Best regards.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions