Cloud Infrastructure Aggregate Representation Classes

Owned by Arun Gupta
Last updated: Jun 05, 2018
4 min read

Introduction to the proposal

Please read the attached Powerpoint as an easy introduction to the proposal.

(Deck updated June 5, 2018: )



Informal Class Diagram 

Aggregate Representation Class Attributes

  • A few key attributes of each aggregation representation class are displayed here. 

  • What is not shown are the various capability, capacity and utilization measures (e.g., CPU capabilities, capacity, optional allocation step size, and utilization) that are to be attached the classes. These measures are essential for placement and management purposes in Edge Automation.  These will be added once a consensus is reached on the overall class structure.

Physical DC Endpoint Class:

ID

Type

Cardinality

Description

Common-name

Name class

1

Common name fields across all objects

Longitude-ID

float

1

Physical DC Longitude.

Latitude-ID

float

1

Physical DC Latitude. Lat/Lon distance between physical DCs serves as a good estimate for propagation latency.

Network-Bandwidth-Info

(DC common-name, value)

1..n

Maximum incoming/outgoing Network Bandwidth from the physical DC to all other interconnected physical DCs.

Cloud Region Class

Key Attributes

ID

Type

Cardinality

Description

Common-name

Name class

1

Common name fields across all objects

Physical-DC-Endpoint-Collection-ID

Physical DC Endpoint Class Collection

1..n

Used for latency and bandwidth accounting across physical DCs in a distributed data center topology

Resource Cluster Group Class

ID

Type

Cardinality

Description

Common-name

Name class

1

Common name fields across all objects

Resource-Cluster-Collection-ID

Resource Cluster Class Collection

1..n

Collection of multiple resource clusters

Physical-DC-Endpoint-ID

 Physical DC Endpoint

1

This is primarily useful in a distributed data center topology -- more details in the cloud region section.

Resource Cluster Class



ID

Type

Cardinality

Description

Common-name

Name class

1

Common name fields across all objects

<resource>-Collection-list

<resource> Collection

1..n

e.g., Collection of multiple compute hosts

Resource Slice Class

ID

Type

Cardinality

Description

Common-name

Name class

1

Common name fields across all objects

<tenant>-name

Name class

1

Reference to the <tenant>/administrative domain to whom the slice is given.

<allocated resources>-list

<resource-allocation> Collection

1..n

Resources with allocations

Important Differences between Public Cloud and Private Cloud

Private cloud offers more fine grained control over the infrastructure as compared to Public Cloud

  • Public cloud exposes only Virtualized infra layer

  •  

    • Virtualized infra layer objects

    •  

      • Aggregate object example: Resource Slice

      • Atomic object example: VM

  • Private cloud exposes HW infra layer besides Virtualized infra layer

  •  

    • Virtualized infra layer objects (same as Public Cloud) +

    • HW infra layer objects

    •  

      • Aggregate object example: Resource Cluster

      • Atomic object example: Host

Some non-exhaustive examples including benefits of more fine grained HW infrastructure control in a Private Cloud – these are especially relevant for Distributed Edge Clouds

  • Service Security Policy

  •  

    • Leverage Smart NICs to program security policies to deliver performance & scalability

  • Service Operational Policy

  •  

    • Leverage Host and Resource Cluster near-real-time resource metrics and real-time faults & alerts to substantially improve closed loop remediation response time

  • Service Placement Policy

  •  

    • Leverage Resource Cluster near-real-time allocated resource capacity and metrics/faults/alerts to substantially improve dynamic workload placement/scheduling across cloud regions

    •  

      • These can be represented as additional soft constraints in the placement/scheduling minimize/maximize objective function

Appendix A – Usage Examples

Resource Cluster Group Usage Example

VNF Type - EPC CP, PGW DP, SGW DP, BNG DP, IMS CP, etc. – where CP is Control Plane and DP is Data Plane

Resource Cluster A - “RCA” - hosts with standard NICs

Resource Cluster B - “RCB” - hosts with IPSec offload NICs

Resource Cluster Group A - “RCGA” - “RCA” and “RCB”

Resource Cluster Group B - “RCGB” - “RCB”

Resource Slice A - “RSA” - logical slice in “RCGA”

Resource Slice B - “RSB” - logical slice in “RCGB”

Resource Slice C - “RSC” - logical slice in “RCGB”

Realizing Minimum guarantees

  • Sum of minimum guarantees of resource slices cannot exceed the total capacity of the resource clusters they belong to

Placement Policy Example

  • VNF Type “EPC CP” uses “RSA”

  •  

    • NIC offloads are immaterial to EPC CP; EPC CP can use “RCA” and “RCB”

  • VNF Type “PGW DP” uses “RSB”

  •  

    • IPSec offload in NIC is a must for PGW DP performance reasons



VNF Type Examples

Control Plane (CP)                             – MME, 5G CU-CP

Data Plane (DP) or User Plane (UP)  - PGW, SGW, BNG, 5G CU-UP

VNF Type maps to Resource Cluster Group Class (a Resource Cluster Group could have one or more Resource Clusters)

  • Multi Cloud Mapping of Resource Cluster Group

    • <Host Aggregate> in OpenStack; <Host Aggregate, Cluster> in VMware integrated OpenStack

vCPE Optimization Policy Example using Aggregate Objects

R1 vCPE use case – Illustrative Sequence Diagrams (https://wiki.onap.org/display/DW/Residential+Broadband+vCPE+Drafts+for+discussion?preview=%2F10783327%2F16005563%2FvCPE+Use+Case+-+Customer+Service+Instantiation+-+171103.pptx)

Constraints used by Optimization Framework (OOF)

  • VBNG location is fixed based on subscriber

  • VG MUX to VBNG Data Center connectivity latency cannot exceed certain value

Optimization Policy used by OOF

  • Choose optimized multi cloud instance for the placement of VG MUX for a given subscriber based on the above  

VNF Mapping to Infrastructure

  • In this example, each VNF maps to a Resource Slice in the infrastructure

  •  

    • Multi Cloud Mapping of Resource Slice

    •  

      • <Project> in OpenStack and VMware Integrated OpenStack; <Tenant> in Azure

Multi Cloud inputs used by OOF

  • Near-real-time stats per <Resource Slice, Resource Cluster Group> at scale using asynchronous push model using DMaaP

5G Optimization Policy Example using Aggregate Objects

R3 5G use case – Definition, Creation & Management of Network Slices; Optimization of the Deployed Network & Slices

Constraints used by Optimization Framework (OOF)

  • 5G CU-UP location is fixed based on group of subscribers

  • 5G CU-CP to 5G CU-UP Data Center connectivity latency cannot exceed certain value

Optimization Policy used by OOF

  • Choose optimized multi cloud instance for the placement of 5G CU CP for a group of subscribers based on the above  

VNF Mapping to Infrastructure

  • In this example, each VNF maps to a Resource Slice in the infrastructure

  •  

    • Multi Cloud Mapping of Resource Slice

    •  

      • <Project> in OpenStack and VMware Integrated OpenStack; <Tenant> in Azure

  • For the 5G CU-UP Data Plane VNF type, each VNF maps to a specific "Physical DC Endpoint" in a "Cloud Region"

Multi Cloud inputs used by OOF

  • Near-real-time stats per <Resource Slice, Resource Cluster Group> at scale using asynchronous push model using DMaaP