Exclusive Port-Pair Group for Non-Transparent Service Functions

URL of the launchpad blueprint:


This specification describes the support for non-transparent Service Functions in SFC Port Chains using a SFC Port Pair Group that is used exclusively by one Port Chain. Non-transparent Service Functions modify the N-tuple header fields of a packet.

Problem Description

Most legacy Service Functions (SF) do not support SFC encapsulation, such as NSH, and therefore require an SFC Proxy to re-classify a packet that is returned from the egress port of the SF. The SFC Proxy uses the N-tuple values of a packet header to re-classify a packet. The packet N-tuple consists of the following:

  • Source IP address

  • Destination IP address

  • Source TCP/UDP port

  • Destination TCP/UDP port

  • IP Protocol

However, if the SF is non-transparent (it modifies a part of the N-tuple of a packet), then re-classification cannot be done correctly. See https://datatracker.ietf.org/doc/draft-song-sfc-legacy-sf-mapping/

In addition the SF may dynamically change the mapping of the N-tuple values as the SF operations progress. A mechanism that uses a static N-tuple mapping to adjust for N-tuple changes cannot be employed.

Proposed Changes

This is an enhancement to the SFC proxy so that it can handle the dynamic changes to N-tuple translation rules of the SF.

A solution to the non-transparent SF is to use a SF VM that has multiple instances and assign the port-pairs for each SF instance to a separate Port Chain.

This can be done by adding these ports to a SFC Proxy Port Pair Group which operates as a Port Pair Correlation Map instead of a normal Load Distribution function. The Proxy Port Pair Group is configured with multiple Port Pairs that are attached to the SF Instances of a specific non-transparent SF type, such as a Firewall SF. This Port Pair Group is configured to operate as a Port Pair Correlation Map.

Each non-transparent SF instance is attached to a single Port Pair. These SF instances may either run on a VM or on a container within a VM. If an SF instance runs within a container, the container sub-port ([1][2]) is used as the ingress and/or egress port of the Port Pair.

Each Port Chain is mapped to one of these port-pairs. Packets for a Port Chain arriving at the OVS Integration bridge are steered to the ingress port of the Port Pair assigned to that Port Chain. Packets received back from the SF on its egress port are then mapped back to the corresponding Port Chain. This mechanism avoids the need for the SFC Proxy to re-classify packets returned from the egress port of the non-transparent SF.

For example, in the figure below, packets on Port Chain A are steered to Port Pair 1 and sent to the ingress port of SF Instance 1. Packets from the egress port of SF Instance 1 are then mapped back to Port Chain A and are delivered to the next hop in the chain.

When a Port Chain is created (or updated) that uses a SFC Proxy PPG, the Port Chain is assigned to one of the Port Pairs in the PPG and the Port Pair is reserved for that Port Chain. If the Port Chain is deleted or the PPG is removed from the Port Chain, its Port Pair becomes available for use by another Port Chain.

The Port Pairs in the SFC Proxy Port Pair Group may be hosted on different Compute Nodes as shown in the diagram below.

If a Port Chain is created that uses a SFC Proxy Port Pair Group and all the Pairs in that PPG are in use by other Port Chains, an error ‘Maximum number of Port Chains reached’ is returned.

This obviously requires that multiple instances of the non-transparent SF be deployed in either VMs or containers. The number of SF instances that must be deployed and configured as Port Pairs depends on the maximum number of Port Chains that are expected to use that particular SF. However, deploying multiple instances of a SF is easily done in modern data centers.

A Port Chain may include multiple SFC Proxy PPGs, each one for a different type of non-transparent SF. For example PPG1 may be a group of non-transparent Firewall SF instances and PPG2 may be a group of non-transparent HTTP Optimizer SF instances.

        Compute Node 1
|                                                            |
|     OVS Integration Bridge          Non-transparent SF     |
|  +--------------------------+  +.........................+ |
|  |   SFC Proxy Port Pair    |  .                         . |
|  |   Correlation Map PPG    |  .        VM/Container1    . |
|  | +.....................+  |  . pp1+------------------+ . |
|  | .Port Chain A <-> pp1 .--------->| Non-transparent  | . |
|  | .                     .<---------|  SF Instance  1  | . |
|  | .                     .  |  .    +------------------+ . |
|  | .                     .  |  .        VM/Container2    . |
|  | .                     .  |  . pp2+------------------+ . |
|  | .Port Chain C <-> pp2 .--------->| Non-transparent  | . |
|  | .                     .<---------|  SF Instance 2   | . |
|  | .                     .  |  .    +------------------+ . |
|  +-.---------------------.--+  .                         . |
     .  Compute Node 2     .     .                         .
|    .                     .     .                         . |
|    .OVS Integration Bridge     .                         . |
|  +-.---------------------.--+  .                         . |
|  | .                     .  |  .        VM/Container3    . |
|  | .                     .  |  . pp3+------------------+ . |
|  | .Port chain X <-> pp3 .--------->| Non-transparent  | . |
|  | .                     .<---------|  SF Instance 3   | . |
|  | +.....................+  |  .    +------------------+ . |
|  |                          |  +.........................+ |
|  +--------------------------+                              |


An alternative mechanism for non-transparent SFs is to mark PPG as exclusive so that it is assigned to one port chain only. This would require a PPG be created for each port chain. The advantage to this approach is that the PPG can be used for load balancing.

Data model impact

Add a “proxy-correlation-map” attribute to the Port Pair Group. This is a Boolean that will enable the Proxy Port Correlation. Add an “exclusive” attribute to the Port Pair Group. This is a Boolean that will enable exclusive use of a Port Pair Group by one Port Chain.

REST API impact

Add “proxy-correlation-map”: true to the Port Pair Group. Add “exclusive”: true to the Port Pair Group.

Security impact


Notifications impact


Other end user impact


Performance Impact


Other deployer impact


Developer impact




Work Items

1. Extend API port-pair-group-parameter to support “proxy-correlation-map” and the “exclusive” attributes. 2. Extend networking-sfc OVS driver to support “proxy-correlation-map” and “exclusive” attributes. 3. Add unit and functional tests. 4. Update documentation.




Unit tests and functional tests will be added.

Documentation Impact



[1] Neutron Trunk-port https://wiki.openstack.org/wiki/Neutron/TrunkPort

[2] VLAN aware VMs https://review.openstack.org/#/c/243786/11/specs/mitaka/vlan-aware-vms.rst