Alcatel-Lucent supports Hybrid OpenFlow Switch (H-OFS) functionality. The hybrid model allows operators to deploy Software Defined Network (SDN) traffic steering using OpenFlow (OF) atop of the existing routing/switching infrastructure.
Topics in this chapter include:
The hybrid OpenFlow model allows operators to deploy Software Defined Network (SDN) traffic steering using OpenFlow atop of the existing routing/switching infrastructure. Some of the main benefits of the hybrid model include:
In a basic mode of operation, a single OpenFlow Switch instance is configured on the router and controlled by a single OpenFlow controller.
The OF controller(s) and router exchange OpenFlow messages using the OpenFlow protocol (version 1.3.1) over the TCP/IPv4 control channel. Both out-of-band (default) and in-band management is supported for connectivity to the controller. An OpenFlow message is processed by the OpenFlow switch instance on the router that installs all supported H-OFS traffic steering rules in a flow table for the H-OFS instance. A single table per H-OFS instance is supported initially.
The H-OFS allows operators to:
Steering actions programmed using OpenFlow are functionally equivalent to ACL actions. Please see later sections for more details on how OpenFlow standard messages are translated by the SR OS OpenFlow switch into SR OS ACL filter actions.
The router allows operators to control traffic using OF, as follows:
To enable rules in a given H-OFS on an existing service router interface, an operator must:
OpenFlow can be embedded in IPv4/IPv6 ACL filter policies deployed on:
OpenFlow functionality is supported in addition to all existing functionality on a given interface and can be enabled with no impact on forwarding performance. Operators can move from CLI/SNMP programmed steering rules to OpenFlow operational model in service without service disruption.
The operator can configure one or more instances of an H-OFS (using SNMP/CLI interfaces) with each instance controlled by an OF-controller over a unique OF channel (using openflow protocol). One OF controller can control multiple H-OFS instances (using dedicated channels), or a dedicated OF controller per switch can be deployed. For each switch, up to two OF controllers can be deployed for redundancy. If two controllers are programmed, they can operate in either OFPCR_ROLE_EQUAL roles or in OFPCR_ROLE_MASTER and OFPCR_ROLE_SLAVE roles. Figure 27 depicts this architecture:
SR OS supports two modes of operations for an H-OFS instance: GRT-only and multi-service. The mode of operations is operator-controlled per H-OFS instance by enabling or disabling switch-defined-cookie option (configure>open-flow>of-switch>flowtable 0). For backward compatibility, GRT-only mode of operation is default but, since multi-service mode is a functional superset, it is recommended to operate in multi-service mode whenever possible. The operator can change the mode in which an H-OFS instance operates but a shutdown is required first. This will purge all the rules forcing the OF controller to reprogram the switch instance once re-enabled in a new mode. An SR OS router supports both H-OFS operational modes concurrently for different switch instances.
Multi-service operational mode uses part of the FlowTable cookie field (higher order 32 bits) to provide the enhanced functionality; the lower order FlowTable cookie bits are fully controlled by the OF controller. Table 65 depicts higher order bit Flow Table cookie encoding used when operating in the multi-service mode.
sros-cookie Name | sros-cookie Type (Bits 63...60) | sros-cookie Value (Bits 59...32) | FlowTable Entry Interpretation Based on the sros-cookie |
grt | 0000 | 0 | FlowTable rule is applicable to GRT instance (IES and router interfaces) |
system | 1000 | 0 | FlowTable rule is applicable to system filters |
service | 1100 | service-id for existing VPLS or VPRN service | FlowTable rule is applicable to an existing VPRN or VPLS service specified by the sros-cookie value |
To enable multi-service mode of operation, an operator must embed the OF switch in an ACL filter policy, and, since multi-service H-OFS supports a mix of VPRN/VPLS/GRT/System rules, an additional scope of embedding must be selected (embed open-flow service, embed open-flow system - grt scope used by default). Since after embedding H-OFS instance, an ACL policy contains rules specific to a particular VPRN or VPLS service instance or to a GRT or to a System Filter Policy, the ACL filter policy can only be used in the scope defined by H-OFS embedding.
Rules programmed by an OF controller with grt, system, and service cookies specified are accepted even if the H-OFS instance is not embedded by a filter activated in a given context. Rules programmed by an OF controller with a service cookie specified, when the service ID is not one of the supported service types, or when the service with the specified id does not exist, are rejected with an error returned back to the controller. If an H-OFS is embedded into a line card policy with a specific service context, the embedding must be removed before that service is deleted.
Table 66 summarizes the main differences between the two modes of operation.
Function | GRT Mode (no switch-defined-cookie) | Multi-service Mode (switch-defined-cookie) |
Support OF on IES access interfaces | Yes | Yes |
Support OF on router interfaces in GRT instance | Yes | Yes |
Support OF on VPRN access and network interfaces | No (lack of native OF service virtualization) | Yes |
Support OF on VPLS access and network interfaces | No (lack of native OF service virtualization) | Yes |
Support port and VLAN in flowtable match (see the following section) | No | Yes |
Support OF control of System ACL policies | No | Yes |
Traffic steering actions | Forward, drop, redirect to LSP, Layer 3 PBR actions only | All |
Scale | Up to ingress ACL filter policy entry scale | Up to OF system scale limit per H-OFS instance, and up to 64534 entries per unique sros-cookie value |
Caveats:
When operating in multi-service mode, SR OS H-OFS supports matching on port and VLAN IDs as part of Flow Table match criteria. When an OF controller specifies incoming port and VLAN values other than "ANY", the H-OFS instance translates them to an SR OS VPLS SAP (sros-cookie must be set to a valid VPLS service ID). If the translation does not result in an existing VPLS SAP, the rule is rejected and an error is returned to the controller.
A flow table rule with a port/VLAN ID match is programmed only if the matching SAP has this H-OFS instance embedded in its ACL ingress filter policy using SAP scope of embedding (embed open-flow sap). Please see SR OS H-OFS Port and VLAN Encoding for required encoding of port and VLAN IDs.
The SR OS H-OFS supports a mix of rules with service scope and with SAP scope. For VPLS SAPs, an H-OFS instance must be embedded twice: once for the VPLS service and once for the SAP if both service-level and SAP-level rules are to be activated.
An example of activating both service-level and SAP-level rules inside a single ACL policy 1 used on VPS SAP 1/1/1:100:
Caveats:
A router H-OFS instance is embedded into line card IPv4 and IPv6 filter policies to achieve OF-controlled Policy Based Routing (PBR). When H-OFS instance is created, embedded filters (IP and IPv6) required for that instance are automatically created. The filters are created with names, as follows:
“_tmnx_ofs_<ofs_name>”, with the same name for IPv4 and IPv6 filters used.
If embedded filters cannot be allocated due to the lack of filter policy instances, the creation of an H-OFS instance will fail. When the H-OFS instance is deleted, the corresponding embedded filters are freed.
The H-OFS can be embedded only in ingress filter policies on line cards/platforms supporting embedded filters (FP2-based or newer) and for services supporting H-OFS. Embedding of an H-OFS in filter policies on unsupported services is blocked, embedding of an H-OFS in filter policies in unsupported direction or on unsupported hardware follows the general filter policy misconfiguration behavior and is not recommended. Unsupported match fields are ignored. Other match criteria may cause a packet to match an entry.
As soon as an H-OFS instance is created, the controller can program OF rules for that instance. For instance, the rules can be created prior to the H-OFS instance embedding into a filter policy or prior to a filter policy with H-OFS instance embedded being assign to an interface. This allows operator to either preprogram H-OFS steering rules, or to disable the rules without removing them from a flow table by removing the embedding. An error is returned to controller if it attempts to program rules not supported by the system. The following lists examples of the errors returned:
As the OF controller updates traffic steering rules, the Hybrid OpenFlow Switch updates the flow table rules. This automatically triggers programming of the embedded filter, which consequently causes instantiation of the rules for all services/interfaces that have a filter policy embedding this H-OFS instance. Embedding filter policy configuration/operational rules apply also to embedded filters auto-created for an H-OFS instance (see Embedded Filter Support for ACL Filter Policies section of this guide). MPLS cannot be deleted if OFS rules are created that redirect to an LSP.
The auto-created embedded filters can be viewed through CLI but cannot be modified and/or deleted through filter policy CLI/SNMP. Operator can see the above embedded filters under show filter context, including the details on the filters themselves, entries programmed, interface association, statistics, etc.
The following picture depicts the H-OFS to service operator-configurable mapping example.
For an H-OFS with switch-defined-cookie enabled, embedded filters are created for each unique context in the H-OFS instead.
The router allows mixing H-OFS rules from one or more H-OFS instances in a single filter policy. Co-existence of H-OFS rules in a single policy with CLI/SNMP programmed rules and/or BGP flowspec programmed rules in a single line card filter policy is also supported. When a management interface and an OF controller flow entry have the same filter policy entry, the management interface-created entry overrides the OF controller-created entry; see the embedded filter functional description. For mixing of the rules from multiple management entities, the controller should not program an entry in its Flow Table that would match all traffic, as this would stop evaluation of the filter policy.
The router supports HA for the OF Flow Table content and statistics. On an activity switch the channel goes down and is re-established by the newly active CPM. “Fail secure mode” operation takes place during channel re-establishment (OpenFlow rules continue to be applied to the arriving traffic). OF controller is expected to re-synchronize the OF table when the channel is re-established.On a router reboot, H-OFS Flow Table rules and statistics are purged. The same takes place when H-OFS instance is shutdown. The H-OFS instance cannot be deleted unless the H-OFS instance is removed first from all embedding filter policies.
SR OS Hybrid OpenFlow switch supports statistics retrieval using the OpenFlow protocol. There are two types of statistics that can be collected:
Operational Notes:
The H-OFS supports auxiliary channels, as defined in OpenFlow version 1.3.1. The packet-in and statistics functions are supported on the auxiliary channels as well as on the main channel. When the auxiliary channel is enabled on a switch (using the aux-channel-enable command), the switch will set up a dedicated auxiliary channel for statistics (Auxiliary ID 1) and a dedicated auxiliary channel for packet-in (Auxiliary ID 2) if a packet-in action is configured, to every controller for a given H-OFS switch instance. Auxiliary connections use the same transport as the main connection. The switch will handle any requests over any established channel and respond on the same channel even if a specific requested auxiliary channel is available. The H-OFS instance uses the packet-in connection for packet-in functionality by default and expects (but does not require) the controller to use the statistics channel for statistics processing by default. The switch uses the auxiliary channels (packet-in for packet-in-specific requests and statistics for statistics-specific requests) as long as they are available. If they are not available, the switch will use the next available auxiliary channel. If none of the auxiliary channels are available, the main channel will be used.Auxiliary connections can be enabled or disabled without shutting down the switch.
As described in the previous section, an update to an OpenFlow Switch’s flow table, results in the embedded filter update(s), which triggers update to all filter policies embedding those filters. The router automatically downloads the new set of rules to the line cards as defined through service configuration. The rules became part of ingress line card pipeline as depicted in Figure 29.
Logical ports are used in OpenFlow to encode switch-specific ports. SR OS H-OFS uses logical ports in steering actions by encoding PBR targets. The following encoding shows logical port types supported by SR OS H-OFS:
Bits 31..28 | Bits 27..24 | Bits 24..0 |
Logical port type (LPT) | Logical port type sub-type (LPT-S) | Logical port type value (LPT-V) — always padded with leading zeros |
Encodings:
The supported range in OF is limited to a 24-bit service ID value range (a subset of VPRN IDs supported by the SR OS system).
Logical port values other than RSVP-TE LSP and MPLS-TP LSP require H-OFS with switch-defined-cookie enabled. GRT instance and VPRN ID logical ports are not stored in the H-OFS logical port table, hence functionality such as retrieving statistics per port is not available for those values.
The OF controller can use port and VLAN values other than ANY for VPLS SAP match and for VPLS steering to SAP for H-OFS instances with switched-defined-cookie enabled.
To specify a port in an OF message, SR OS TmnxPortId encoding must be used. The allowed values are those for Ethernet physical ports and LAG.
To encode VLAN tags, OXM_OF_VLAN_ID and new experimenter OFL_OUT_VLAN_ID fields are used as per Table 68.
NULL tag, dot1Q tag, inner QinQ tag VlanId | Outer QinQ tag VlanId |
OXM_OF_VLAN_VID | OFL_OUT_VLAN_ID (Experimenter field uses same encoding as OXM_OF_VLAN_VID) |
Table 69 shows how OF programmed values are translated to SR OS SAPs.
OXM_OF_IN_PORT | OXM_OF_VLAN_VID | OFL_OUT_VLAN_ID | Matching SAP SR OS Encoding | Supported in flow_add | Supported in flow_mod flow_del mp_req | Comment |
TmnxPortId for port or LAG | Value: 0x0000 Mask: Absent | Must be absent | port-id lag-id | Yes | Yes | Mask must be absent |
TmnxPortId for port or LAG | Value: 0x1yyy, yyy encodes qtag1 Mask: Absent | Must be absent | port-id:qtag1 lag-id:qtag1 | Yes | Yes | Mask must be absent |
TmnxPortId for port or LAG | Value: 0x1FFF Mask: Absent | Must be absent | port-id:* lag-id:* | Yes | Yes | Mask must be absent |
TmnxPortId for port or LAG | Value: 0x1000 Mask: 0x1000 | Must be absent | port-id: any lag-id: any where "any" is either * or a valid VLAN-ID (but not NULL) | No | Yes | Mask must be 0x1000 |
TmnxPortId for port or LAG | Value: 0x1yyy, yyy encodes qtag2 Mask: Absent | Value: 0x1zzz, zzz encodes qtag1 Mask: Absent | port-id:qtag1.qtag2 lag-id:qtag1.qtag2 | Yes | Yes | Mask must be absent |
TmnxPortId for port or LAG | Value: 0x1FFF Mask: Absent | Value: 0x1zzz, zzz encodes qtag1 Mask: Absent | port-id: qtag1.* lag-id: qtag1.* | Yes | Yes | Mask must be absent |
TmnxPortId for port or LAG | Value: 0x1FFF Mask: Absent | Value: 0x1FFF Mask: Absent | port-id: *.* lag-id: *.* | Yes | Yes | Mask must be absent |
TmnxPortId for port or LAG | Value: 0x1000 Mask: 0x1000 | Value: 0x1zzz, zzz encodes qtag1 Mask: Absent | port-id: qtag1.any lag-id: qtag1.any where any is either * or a valid VLAN-ID (but not NULL) | No | Yes | Mask must be absent for OFL_OUT_VLAN_VID |
TmnxPortId for port or LAG | Value: 0x1000 Mask: 0x1000 | Value: 0x1FFF Mask: Absent | port-id: *.any lag-id: *.any where "any" is either * or a valid VLAN-ID (but not NULL) | No | Yes | Mask must be absent for OFL_OUT_VLAN_VID |
TmnxPortId for port or LAG | Value: 0x1000 Mask: 0x1000 | Value: 0x1000 Mask: 0x1000 | port-id: any.any lag-id: any.any where "any" is either * or a valid VLAN-ID (but not NULL) | No | Yes | Masks must be 0x1000 |
TmnxPortId for port or LAG | Value: 0x0000 Mask: Absent | Value: 0x1FFF Mask: Absent | port-id: *.null | Yes | Yes | Mask must be absent |
A router supports redirection of IPv4 or IPv6 next-hop for traffic arriving on a L3 interface. An OF controller can rely on this functionality and program PBR next-hop steering actions for H-OFS instances with switched-defined-cookie enabled using the following OF encoding:
In case of erroneous programming, the following experimenter-specific errors are returned to the controller:
A router supports redirection of IPv4 or IPv6 traffic arriving on an L3 interface to a different routing instance (GRT or VRF). An OF controller can rely on this functionality and program PBR actions for GRT/VRF steering for H-OFS instances with switched-defined-cookie enabled using the following OF encoding:
port= SR OS LOGICAL port encoding GRT or VPRN Service ID as outlined in the SR OS H-OFS Logical Port section.
Since a 24-bit value is used to encode the VPRN service ID in the logical port, redirection to a VPRN service with a service ID above that range is not supported.
A router supports redirection of IPv4 or IPv6 traffic arriving on an L3 interface to a different routing instance (GRT or VRF) and next-hop IP at the same time. An OF controller can rely on this functionality and program PBR steering action for H-OFS instances with switched-defined-cookie enabled using the following OF encoding:
Encoding as outlined in the Redirect to IP next-hop section (indirect flag must be set).
port= SR OS LOGICAL port encoding GRT or VPRN Service ID as outlined in the SR OS H-OFS Logical Port section.
The router supports traffic steering to an LSP. The following details the OF encoding to be used by an OF controller:
port= SR OS LOGICAL port encoding RSVP-TE or MPLS-TP LSP as outlined in SR OS H-OFS Logical Port section
A received LSP in a flow rule is compared against those in the H-OFS logical port table, if the table does not contain the LSP the rule programming fails. Otherwise, the rule is installed in an ACL filter. As long as any path within the LSP is UP, the redirect rule will forward unicast IP(v6) traffic on the currently used best LSP path by adding LSP transport label and, in case of IPv6 traffic, additionally adding explicit NULL label.
When an LSP in the H-OFS logical port table goes down, the OF Switch removes the LSP from its logical port table and may notify the controller of that fact if the logical port status reporting is enabled. It is up to the OF controller to decide whether to remove rules using this LSP or not. If the rules are left in the flow table, the traffic that was to be redirected to this LSP will instead be subject to a forward action for this flow rule. If the controller does not remove the entries and the system re-uses the LSP identified for another LSP, the rules left in the flow table will start redirecting traffic onto this new LSP.
In some deployments, an SDN controller may need to learn from the router H-OFS logical ports status. To support that function, the OF switch supports optional status reporting using asynchronous OF protocol messages for ports status change.
For traffic arriving on a VPLS interface, a router supports PBF to steer traffic over another VPLS SAP in the same service. An OF controller can rely on this functionality and program PBF steering action for H-OFS instances with switched-defined-cookie enabled using the following OF encoding:
port: = Encoding as outlined in SR OS H-OFS Port and VLAN Encoding section
OXM TLVs encode SAP VLANs as outlined in SR OS H-OFS Port and VLAN Encoding section:
For traffic arriving on a VPLS interface, a router supports PBF to steer traffic over a VPLS SDP in the same service. An OF controller can rely on this functionality and program PBF steering action for H-OFS instances with switched-defined-cookie enabled using the following OF encoding:
In case of erroneous programming, the following experimenter-specific errors are returned to the controller:
An OF controller can program forward action, when a specific flow is to be forwarded using regular router forwarding. This would be a default behavior if the filter-policy embedding this OF switch instance has a default-action forward and no filter policy rule matches the flow. To implement forward action, the following OF encoding is used:
where NORMAL is a OF reserved value.
An OF controller can program a drop action, when packets of a specific flow are to be dropped. To implement drop action, the following OF encoding is used:
Packets that do not match any of the flow table entries programmed by the controller are subject to a default action. The default action is configurable in the CLI using the no-match-action command. Three possible no-match actions are supported: drop, fall-through (packets are forwarded with regular processing by the router), and packet-in.
The packet-in action causes packets that do not match entries in the flow table as programmed by the OpenFlow controller to be extracted and sent to the controller in a flow-controlled manner. Because EQUAL is supported, packet-in messages are sent to all controllers in the UP state. Only the first packet of a given 5-tuple flow (source IP address, destination IP address, source port, destination port, protocol) to which the no-match action is applied is sent to the controller in order to protect the controller. This 5-tuple flow context ages out after 10 s. Each switch instance maintains contexts for up to 8192 outstanding packet-in messages to the controller. If the packet-in action is used, an auxiliary channel should be enabled for packet-in processing (using the aux-channel-enable command). A count of packets to which packet-in is applied is also available through the OpenFlow channel statistics.
The following information describes OF implementation caveats: