Enable scaling function for VNFFG

Enable scaling function for VNFFG

https://blueprints.launchpad.net/tacker/+spec/vnffg-scaling

This proposal aims to provide scaling function for VNFFG based on the existing policy actions in VNFM. The spec is referred to Network service fault management in ETSI standard [1].

Problem description

Currently, Tacker has already supported scaling function for individual VNFs by using alarm driver. However, when doing scaling in or out, the new-added or deleted instances’ CP cannot be added or removed to/from Neutron SFC port-pair group.[#second]_

This spec is supposed to provide scaling for VNFFG in single site, multi-sites support will be taken into account in the future. In addtition, Tacker had plan to add VNFFG to Network Service (NS), therefore this spec also considers extending scope to support VNFFG inside NSs.

Proposed change

Introduce a vnffg-ha engine in Tacker NFVO

+---------------------------------+
|          Tacker_NFVO            |
|      +------------------+       |
|      |                  |       |
|      |  NFVO API/TOSCA  |       |               +-----------------------+
|      |                  |       |               |                       |
|      +---------+--------+       |               |      Tacker_VNFM      |
|                |                |               | +-------------------+ |
|                |                |               | |  VNFM API/TOSCA   | |
|                |                |               | |                   | |
|                |                |               | +---------+---------+ |
|                |                |               |           |           |
|                |                |               |           |           |
|                |                |               |           |           |
|                |                |               |           |           |
|          +-----v-----+          |               |    +------v------+    |
|          |NFVO Plugin|          |               |    | VNFM Plugin |    |
|          |           |          |               |    |             |    |
|          +-----+-----+          |               |    +------+------+    |
|                |                |               |           |           |
|  +-------------v--------------+ |               |           |           |
|  |      vnffg-ha engine       | |               |   +-------v--------+  |
|  |                            <-----conductor-------+ policy actions |  |
|  |                            | |               |   +----------------+  |
|  +----------------------------+ |               +-----------------------+
|                                 |
+---------------------------------+

Component

vnffg-ha engine: When vnffg-ha receives a trigger from policy actions in VNFM:
  1. Firstly, it finds which VNFFG VNF belongs to.
  2. Then it makes changes in VNFFG corresponding to VNF policy action.

Scaling support for VNFFG in Neutron SFC [4]:

             +--------------------------------------------------+
             |     +---------------+  Neutron SFC               |
             |     |  +---------+  |             +---------+    |
    +-----------------+         +----------------+         |    |
    |        |     |  | VM 11   |  |             | VM 2    |    |
+---+----+   |     |  +---------+  |             +-----+---+    |
|Endpoint|   |     |               |                   |        |
|        |   |     |  +---------+  |                   |        |
+--------+   |     |  |         |  |                   |        |
    +-----------------+ VM 12   +----------------------+        |
             |     |  +---------+  |                            |
             |     +---------------+                            |
             +--------------------------------------------------+

In the above figure, VM11 is launched by VNF1, VM12 is scaled out from VM11. VM2 is launched by VNF2.

Currently, Tacker leverages networking-sfc project [3] to deploy SFC based on Neutron ports. One of good points is that networking-sfc enables multiple port pairs inside a port-pair group [4] so that load-balancing could be applied.

By using port pair group update we can modify the lists of port pairs including add, remove, or even change the new set of port pairs. This feature in networking-sfc is feasible to do auto-scaling for VNFFG.

Load-balancing

There are several cases to process: 1. load-balancing between VNFs. 2. load-balancing between VDUs

In the scope of this spec, we deal with load-balancing between VDUs to achieve scaling vnffg.

In Neutron-SFC, a logical port-pair-group can contain one or more logical port-pairs and is used to load balance traffic across the Service Functions (logical port-pairs) [6]. We will use this spec to perform scale-out or scale-in operations by adding or removing port-pairs on a port-pair-group for VNFFG [7].

For example, insert port-pair (PP2) of VM12 to existing port-pair-group that contain port-pair (PP1) of VM11 to perform scale-out.

$ neutron port-pair-group-update --port-pair PP1 --port-pair PP2 PPG1

In the same way, we can update port-pair-group to perform scale-in by removing one or more port-pairs from the port-pair-group.

Proposed change

AutoScalingRPC call

class AutoScalingRPC(object):

   target = oslo_messaging.Target(
       exchange='vnffg-scaling',
       topic=topics.TOPIC_CONDUCTOR,
       fanout=False,
       version='1.0')

   def vnf_scaling_event(self, context, **kwargs):
       pass

Tosca template:

tosca_definitions_version: tosca_simple_profile_for_nfv_1_0_0
description: Demo example

metadata:
  template_name: sample-tosca-vnfd1

topology_template:
  node_templates:
    VDU1:
      type: tosca.nodes.nfv.VDU.Tacker
      capabilities:
        nfv_compute:
          properties:
             num_cpus: 1
             mem_size: 512 MB
             disk_size: 1 GB
  properties:
    image: cirros-0.3.5-x86_64-disk
    availability_zone: nova
    mgmt_driver: noop
    config: |
      param0: key1
      param1: key2
    metadata: {metering.vnf: VDU1}

CP11:
  type: tosca.nodes.nfv.CP.Tacker
  properties:
    management: true
    order: 0
    anti_spoofing_protection: false
  requirements:
    - virtualLink:
        node: VL11
    - virtualBinding:
        node: VDU1

CP12:
  type: tosca.nodes.nfv.CP.Tacker
  properties:
    order: 1
    anti_spoofing_protection: false
  requirements:
    - virtualLink:
        node: VL12
    - virtualBinding:
        node: VDU1

CP13:
  type: tosca.nodes.nfv.CP.Tacker
  properties:
    order: 2
    anti_spoofing_protection: false
  requirements:
    - virtualLink:
        node: VL13
    - virtualBinding:
        node: VDU1

VL11:
  type: tosca.nodes.nfv.VL
  properties:
    network_name: net_mgmt
    vendor: Tacker

VL12:
  type: tosca.nodes.nfv.VL
  properties:
    network_name: net0
    vendor: Tacker

VL13:
  type: tosca.nodes.nfv.VL
  properties:
    network_name: net1
    vendor: Tacker

policies:
- vdu1_cpu_usage_monitoring_policy:
    type: tosca.policies.tacker.Alarming
    triggers:
        vdu_hcpu_usage_respawning:
            event_type:
                type: tosca.events.resource.utilization
                implementation: ceilometer
            metrics: cpu_util
            condition:
                threshold: 50
                constraint: utilization greater_than 50%
                period: 600
                evaluations: 1
                method: avg
                comparison_operator: gt
            metadata: VDU1
            action: [respawn, notify]

In the above template, actions include respawn ** and **notify. Accordingly, respawn action indicates the healing function. Meanwhile, notify action indicates events which are triggered to NFVO layer.

Response to scaling action

For scaling in, we need to remove the terminated instance’s CP from current sfc port-pair group ASAP, to avoid data lose.

But for scaling out, the new instantiated instance may need some configuration before we add it’s CP into sfc port-pair group. This also aims to avoid traffic lose. How to make sure VM is ready to work is a question.

Use cases

  1. Scaling-out

VNFM triggers the scaling out based on policies in VNFD. In VNFM layer, new VNF will be scaled out with the same VNFD like the old one. In NFVO layer, we have 2 options. The first option, after launching new VNF, for auto-scaling VNFFG we will wait and update new VNF’s port-pair to existing port-pair-group, it can take long time to reach normal state due to VM-based VNF. The second option, we can use a algorithm to find the best matched VNF, that have the same VNFD, tenant id and low resource usage and then add its port-pair to existing port-pair-group. The second choice can give lower latency. Scaling-out refers to vertical scale-out use case in [5] IETF draft.

  1. Scaling-in

For scaling-in, first port-pair of VNF will be remove from port-pair-group, then tacker will invoke the scale-in policy to shutdown VNF.

Security impact

Notifications impact

Because the failure of VNFs happened in VNFM layer, VNFFGs is orchestrated in NFVO layer. A broken VNF could make one or several VNFFGs fail. Therefore, we need to have a method to inform VNFFGs about their VNFs. In short term, tacker conductor will be used to emit events from VNFM to NFVO. The future consideration is to use event/auditing functions.

Other end user impact

Performance Impact

None

Other deployer impact

None

Developer impact

None

Implementation

Assignee(s)

Primary assignee:
Tung Doan <doantungbk.203@gmail.com>
Other contributors:

Yan Xing an<yanxingan@cmss.chinamobile>

Phuoc Hoang <hoangphuocbk2.07@gmail.com>

Work Items

  • Implement vnffg-ha engine in NFVO
  • Add API for triggering services in NFVO
  • Modify the existing VNFFG implementation in NFVO plugin
  • Add event/auditing function for vnffg-scaling
  • Add unit and functional tests for vnffg-scaling

Dependencies

None

Creative Commons Attribution 3.0 License

Except where otherwise noted, this document is licensed under Creative Commons Attribution 3.0 License. See all OpenStack Legal Documents.