Add service quality monitoring

[kmohr-soprasteria]

Problem:

• Statistics about the completeness and quality of the PM data shall be provided
• in DPMPD v1.0.*, no such statistics are gathered and exposed
• how to measure completeness and quality

Possible solution/tb discussed:

• specify function(s) to gather (and store) this information
  • granularity?
    • per device/interface combination
    • for the previous day
    • perhaps how many have already been collected today (on demand?))
• data gathering:
  • when/input source?
    • compute some information when resultCC is created/stored in dataStore(batch)?
    • or read everything only from datastore
  • what?
    • if we include normalized timestamps (x:00, x:15, x:30, x:45) we could determine exactly which 15min intervals are missing
    • if we would only have "random" timestamps, we could instead count how many records we have gathered for each device/interface (a counter for each day, where new counts are added)
  • shall this data be stored?
    • if yes, this could be stored in the datastore under the interfaceMetadata
    • depending on what to store
      • e.g. just keep a count or list of timestamps where data was gathered per interface for the previous day
      • or store a matrix, with days as columns and counted records (just a number) as rows

[kmohr-soprasteria]

Proposal

Quality statistics have been identified as needed by customers, but no explicit requirements on what statistics should be provided, have been made.

Legacy - data availability in Mycom

• Mycom had a datav (data availability) monitor
• Mycom has dedicated KPIs that count for each NE (unitIDs, ethernetPorts) how many of the expected records per day have been loaded to Mycom; number of expected records depends on the granularity of measurements
• datav monitor provided dedicated KPIs for NE type (unitId, ethernetPorts) and deviceType
• datav monitor showed aggregated percentages

Example how datav monitor looked like:

What could/should SDN provide?

Possible would e.g. be datav statistics (%) per ... (aggregated over all devices)

• device type
• interface type
• device type & interface type

There could also be a service (possibly for internal usage) to return raw quality statistics, possibly with a list of devices as input.

DataStore extension

The data could be written to an additional place in the dataStore (not to the interface-metadata, because this gets overwritten constantly). Data could be kept for several days, cleanup may be done with an own retentionPeriod parameter.

Proposed schema:

quality-statistics:
  type: object
  x-key: mount-name
  # is a second x-key: deviceType possible?
  required:
    - mount-name
    - device-type
    - interface-list
  properties:
    mount-name:
      type: object
    device-type:
      type: object
    interface-list:
      type: array
      items:
        type: object
        x-key: uuid
        # if second key possible also add layerProtocolName as xkey
        required:
          - uuid
          - layer-protocol-name
          - statistics
        properties:
          uuid:
            type: string
          layer-protocol-name:
            type: string
          statistics:
            type: array
            items:
              type: object
              required:
                - date
                - count
                - count24
              properties:
                date:
                  type: string
                  description: 'The date, as yyyy-MM-dd'
                count:
                  type: integer
                  description: 'The number of 15min records from resultCc seen for the date'
                count24:
                  type: integer
                  description: 'The number of 24h records from resultCc seen for the date'

When/how gather statistics

During iteration through resultCc interfaces in p1IterateAiPmSlices and p1IterateEcPmSlices, the counts per date can be gathered and stored in a temporary variable. After processing has finished the data in the dataStore can be updated with the newly gathered data.

How count:

• for each PM slice determine the seen date
• have a counter per date, which gets increased for each seen slice per date
• when data is added to dataStore and the device/interface is already present from previous resultCc processing, the new counts can be simply added to existing counts.

How to aggregate and provide?

Data in the dataStore is NOT aggregated, but per device/interface level.
Either there needs to be a cyclic process that aggregated the data or this can be done on-demand via a service.
The service could be designed for external or internal use. (Details to be discussed.)

It would also be possible to setup a website, where the quality statistics are shown. Similar to the datav monitor from Mycom. This way consumers could easily check for themselves, how good the data availability currently is.

[openBackhaul]

Specification completed.
Issue could be closed after review completed.