Compute Profile
A compute profile is a "predefined configuration" specifying compute resources such as CPU, GPU, Memory, and storage. Administrators create and publish compute profiles in the PaaS Studio. Once published, these are then immediately visible to end users as SKUs that they can consume via self service.
The lifecycle of Compute Profiles and their configuration is managed by administrative users in the Rafay GPU PaaS Studio. The PaaS studio is an administrative portal purpose built for management of reusable profiles.
Compute profiles are "project scoped" by default i.e. they are locked to the Rafay Project they are created in. As with all resources in a project, they can be optionally shared with selected/all projects.
Info
In addition to the web console, administrators can also perform all the tasks described below programmatically using the platform's REST APIs.
Create Compute Profile¶
To create a compute profile, users should access the PaaS Studio, select the Compute Profiles menu on the left and click on "Add Compute Profile. They will be prompted to specify the following in a wizard type experience.
The compute profile allows the administrator to encapsulate the rules and policies associated with the underlying infrastructure. Details captured in the compute profile are also presented to the end user when they wish to provision instances based on the compute profile.
General Section¶
Unique Name¶
Provide a unique name for the compute profile (e.g. sm-gpu-inst). Note that you cannot change the unique name once the profile has been created.
Display Name¶
This is typically a user friendly name such as Small, Medium, Large that will make sense to the end users. This name is what the end users will see an experience.
Description¶
Provide a multiple sentence description that will help the end users understand more details about the compute profile.
Allocation Type¶
From the drop down, select either "dedicated" or "shared". When you select dedicated, you are indicating in the profile that dedicated resources will be provisioned when a compute instance is spawned using this compute profile.
Dedicated infrastructure may be preferred and ideal for high end model training use cases. Shared infrastructure is ideal for general purpose inference and experimentation use cases.
Environment Template¶
Select the name of the environment template from the drop down. When a compute instance is spawned using the profile, the "infrastructure as code" encapsulated in the associated environment template is used to drive the creation of the necessary infrastructure.
Environment Template Version¶
Select the version of the environment template from the drop down.
Labels¶
Labels are optional metadata (key/value pairs) that will be automatically attached to resulting compute instances when they are launched.
Annotations¶
Labels are optional metadata (key/value pairs) that will be automatically attached to resulting compute instances when they are launched.
End User Configuration¶
These are details that are used to help enhance the end user experience.
Icon URL¶
Provide a URL for the Icon that will be displayed to the end user in the catalog of SKUs.
Readme¶
Provide a detailed readme in plain text or markdown. Administrators can use the readme as a means to educate the end user about the compute profile and optionally even capture instructions.
Input Settings¶
Environment templates backing a compute profile can require/expect inputs from the user. The inputs can either be mandatory/required or optional. This section also allows the PaaS admin to make decisions on whether they would like provide "defaults" and allow end users to "override" it.
Output Settings¶
This section allows the PaaS admin to determine what information they would like to show the end user after the compute instance has been created. These are typically key/value pairs that are dynamically generated by the environment template.
View Compute Profile¶
When PaaS Admins click on the compute profiles menu on the left, the entire list of compute profiles available are listed. They can search for the compute profile by providing input into the search box. Some basic information about a compute profile is shown to the admin in the main listing.
- Unique Name
- Creation Date
- Sharing
- Actions Menu (ellipses)
To view additional details about a compute profile, simply click on it. This will display all the details the administrator had configured during initial creation.
Update Compute Profile¶
Follow the steps above for "View" compute profile to view details about a compute profile. Alternatively, click on the actions (ellipses) on the far right of a compute profile and select "edit". PaaS Admins can now make necessary changes to the compute profile and save it.
Info
Any changes will go into effect only for "new" compute instances that are launched based on the profile. Already running compute instances will not be impacted.
Delete Compute Profile¶
To delete a compute profile, click on the actions (ellipses) on the far right of a compute profile and select delete.
Info
This is a destruction action and cannot be reversed. Once a compute profile is deleted, it is no longer available to end users for use.
Share Compute Profile¶
To share a compute profile with selected or all projects, click on the actions (ellipses) on the far right of a compute profile and select "manage sharing". Now, select how you wish to share the project. The options are:
- None (This is the default)
- All Projects
- Selected Projects
The ability to share compute profiles with other projects allows platform teams to centralize the mgmt of compute profiles and share them hierarchically with downstream projects where they can be used by end users in multiple teams/departments.