The ONAP Data Model reflects the IM's ComputeDesc and MemoryDesc elements as the TOSCA capability type onap.capabilities.Compute.

Basic infrastructure capability type
capability_types:
  onap.capabilities.Compute:
    derived_from: tosca.capabilities.Container
    description: a description of the required hardware resources
    properties:
      num_cpus:
        type: integer
        required: false
        constraints:
          - greater_or_equal: 1
      cpu_frequency:
        type: scalar-unit.frequency
        required: false
        constraints:
          - greater_or_equal: 0.1 GHz
      mem_size:
        type: scalar-unit.size
        required: false
        constraints:
          - greater_or_equal: 0 MB
      storage_size:
        type: scalar-unit.size
        required: false
        constraints:
          - greater_or_equal: 0 MB
      io_bitrate:
        description: bits per second
        type: integer
        required: false
      architecture:
        description: vendor+architecture, for example, Intel64
        type: string
        required: false          
      custom_features:
        description: |
          Additional features description, serialized in a well-known format.
        type: string
        required: false

In addtition to the obvious properties required by the IM, this capability type also includes properties that allow for easy customization:

  • architecture - allows the VFC vendors to focus their requirements on a specific hardware architecture, for example "Intel x86".
  • custom_features - an opaque container for a list of feature definitions. Any text, the only limitation is that it should not break the YAML/TOSCA formatting. The contents of this property is actually not part of the TOSCA model!

In spite the changes made in the capability type defininitions, the definition of the node type onap.nodes.Container  does not change:

onap.nodes.Container 
node_types:
  # the very base of the hierarchy of VDU types
  onap.nodes.Container:
    derived_from: onap.nodes.Resource
    artifacts:
      container_image:
        type: tosca.artifacts.Deployment
        description: an image used to launch the Container
    interfaces:
      Standard:
        start:
          implementation: container_image
    capabilities:
      host:
        type: tosca.capabilities.Container  # the TOSCA Specs type is good enough
        occurrences: [0..UNBOUNDED]
    requirements:
      - host:
          capability: onap.capabilities.Compute:
          occurrences: [0..UNBOUNDED]


Examples of requirement assignments in their simplest form:

Requirement for at least 2 vCPUs 
# .. a node template, omitted...requirements:
  - host:
      node_filter:
        capabilities: onap.capabilities.Compute
        properties:
          num_cpus:
            - greater_or_equal: 2


Examples of requirement assignments in the combined (TOSCA properties + custom_features) form:

Requirement for at least 2 vCPUs + bunch of Intel-specific features 
requirements:
  - host:
      node_filter:
        capabilities: onap.capabilities.Compute
        properties:
          num_cpus:
            - greater_or_equal: 2
          custom_features: |
            {
              "simultaneousMultiThreading": true,
              "simultaneousThreads": 10,
              "logicalCpuPinningPolicy": "Dedicated",
              "logicalCpuThreadPinningPolicy": "Prefer",
              "instructionSetExtensions": ["aes", "sse", "ddio"],
              "directIoAccessToCache": "dca",
              "accelerator": "whatever you like",
              "measuredLaunchEnvironment": "",
              "secureEnclave": "",
              "hypervisorConfiguration": {
                "memoryCompaction": "best",
                "kernelSamePageMerging": true
              },
              "computeRas": "pciDetectedAndCorrectedErrors"
            }

In the example above the text in the custom_features property is actually a JSON document.

The hardware vendors are encouraged to provide a well-defined schema (JSON Schema for JSON-formatted texts, DTD or XML Schema for XML-formatted texts, etc) to govern syntax of their custom_feature texts. For example, a JSON schema for the CPU custom feature fron the example above may look like this:

Example of JSON Schema for CPU custom features 
{
  "definitions": {
	"hypervisorConfiguration": {
		"type": "object",
		"properties": {
			"memoryCompaction": { "enum": [ "best", "worst", "optimal" ] },
			"kernelSamePageMerging": {"type": "boolean" }
		}
	}
  },
  
  "type": "object",
  
  "properties": {
	"simultaneousMultiThreading": {"type": "boolean"},
	"simultaneousThreads": {"type": "integer", "minimum": 1, "maximum": 4},
	"logicalCpuPinningPolicy": {"type": "string" },
    "hypervisorConfiguration": { "$ref": "#/definitions/hypervisorConfiguration" },
	"instructionSetExtensions": {
		"type": "array", 
		"items": {
			"type": "string",
			"enum": [ "aes", "sse", "avx", "cat", "cmt", "mbm", "ddio", "smt", "rdrand" ]
		}
	}
  }
}

The idea of using a schema document together with the custom_features text seems to be very powerful. For a demostration of how such a schema would work, please visit any of the JSON schema validators available online, for example, https://json-schema-validator.herokuapp.com/. Just copy and paste over there the above JSON schema and the value of the custom_feature field and check on the validation results:


TOSCA 1.2 allows using a JSON schema as a TOSCA constraint in parameter definition and node_filter clauses. Using an external schema document (with a Web URL in the TOSCA constraint clause) seems to be especially promising, since such usage combines the benefits of schema-driven validation with the flexibility of having the schema hosted outside:

External JSON schema in a TOSCA requirement assignment 
requirements:
  - host:
      node_filter:
        capabilities: onap.capabilities.infrastructure.CPU
        properties:
          num_cpus:
            - greater_or_equal: 2
          custom_features: 
            - schema: http://my.domain.com/url/datatype.constraints.schema.json




  • No labels

2 Comments

  1. Srinivasa Addepalli

    Can JSON document be part of the CSAR file? 

    Thought it provides flexibility for each CPU vendor to define their own DTD and schema,  isn't there a concern that each CPU vendor define their own format and VNF vendors now need to learn multiple ways to represent hardware feature?

    Is this chosen because TOSCA way of representing HW features challenging?  I have put some of my thoughts on "Hardware Platform Requirements" on representing HW features using TOSCA. Any thoughts on that approach?

    1. Srinivasa Addepalli

      Also, TOSCA functions such as get_input() can't be used in JSON, right?  If so, wouldn't that be a concern?