OAM Processing

Dataflow OAM is the Industry's First Holistic Approach to Data Plane OAM     

Operating public networks requires efficient data plane functions for Operation, Administration and Maintenance (OAM). Links and nodes need to be monitored for protection. Anomalies and faults have to be announced. Access switches need automatic configuration. These network and service level OAM tasks are well defined, for example in the ITU Y.1731 and IEEE 802.1ag standards.

Xelerated's Dataflow OAM provides support to a range of data plane OAM tasks. In addition to Carrier Ethernet and MPLS link and service OAM functionality, the Dataflow OAM functionality includes features for table and hardware maintenance.

In addition to network and service OAM, each node needs pro-active maintenance. Forwarding tables for routing, flow processing and counting are flushed and refreshed. MAC addresses are learnt and aged. The packet processing hardware also requires regular health checks and repair routines need to be carried out to keep memory performing as expected.

Holistic OAM Design

Xelerated has taken a holistic approach to OAM. The concept leverages the flexibility of the Dataflow Architecture present in all Xelerated's programmable packet processing chips. All hardware maintenance operations are programmed as part of the data plane software. This means any amount of the packet processing power can be used for maintenance services, making the design very flexible.

As with forwarding software, the OAM software has two parts divided between the control plane and the data plane. The control plane running on the CPU communicates with the management software and initiates many of the OAM tasks. The data plane software, running on the NPU, is responsible for efficiently carrying out the operations initiated by the control plane.

In the dataflow architecture, forwarding programs are triggered by packets. In a similar manner, OAM tasks are also initiated by packets. The control plane compiles a special type of packet, referred to as Executable Control Message (ECM), and sends it onto the programmable pipeline of the NPU. This then triggers a range of execution tasks as defined by the OAM program.

Dataflow OAM Benefits

There are several benefits to the dataflow architecture approach to OAM:

• Deterministic behavior. OAM traffic is critical for accurate network operations and SLA reporting. The dataflow architecture guarantees accurate OAM traffic generation and monitoring
• Complete CPU offload. Link monitoring and link tracing performed by data plane with no CPU dependence; accurate information provided even if CPU is temporarily not accessible
• Powerful processing. Complete pipeline of processing and classification resources available for OAM program implementation
• Full access to table and engine resources allows for efficient execution of interlinked data plane maintenance tasks
• Table updates are always executed to completion within one operation, securing atomic operations; operations on subsequent data packets will be based on consistent table information.

MAC Table Maintenance

Let us look at one example. MAC learning is a data plane optimization task. When a switch operating in bridge mode learns a source MAC address, the subsequent packets in the opposite direction will be forwarded to the learnt interface only. If the MAC address is unknown the packets will be flooded to all interfaces.

This case well illustrates the strength of the dataflow architecture for node level OAM. When the Ethernet packet arrives, the program will check for a forwarding table entry of the source MAC address. In case this is empty, the program will continue to forward the packet as expected, but in addition, a device register will be set in the execution context to indicate that an ECM message should be compiled based on this packet information. At the end of the pipeline, this register automatically replicates the data of the execution context into an ECM which is looped back to the pipeline. The ECM initiates the MAC learning program which checks the forwarding table entry and updates it with the learnt MAC address, all in one atomic operation. The complete process is performed with no dependency on the CPU, but a replication of the packet may be forwarded to the control plane for MAC table updates of other line cards in chassis-based systems.

Link OAM for Protection Switching

Link monitoring is a fundamental part of packet-based OAM. There are two slightly overlapping standards, one defined in Ethernet in the First Mile IEEE 802.3ah focusing on local link level monitoring, and another by IEEE 802.1ag which allows for virtual link monitoring across multiple network nodes.

Link monitoring put high demands on the data plane. Keep-alive packets (Continuity Check Messages) are generated from each Management End Point (MEP) configured on any node interface. When used for fast reroute (MPLS) or protection switching (Ethernet) these packets are generated every 3.3 millisecond (ms). When the receiving end loses three messages in a row, the link is assumed to be broken. This is registered in 10 ms and if the forwarding table is updated within 40 ms, the network can meet the telecom standard of 50 ms protection switching.

Every CCM message holds up-to-date status information of link and forwarding functions monitored by the MEP. Xelerated's on-chip packet generator compiles a packet based on information received by the CCM control program and sends it off on the target link.

The HX NPUs and the AX programmable Ethernet switches feature several on-chip generators which can be flexibly configured to generate on-chip or off-chip traffic up to 150 mpps. The chips also support automatic generation of CCM messages at a configurable interval.

On the receiving end, the data plane program identifies CCM messages and performs automatic updates to the OAM tables. The tables are scanned by the control plane to check if more than three CCM packets are dropped. In case of link loss, the control plane initiates a forwarding table update to carry out fast protection switching.

Ethernet Service OAM

Service OAM includes standard ways to validate configurations, check connectivity and trigger fault notifications. The IEEE 802.1ag standard offers a framework whereby Ethernet services can be monitored across service provider domains. This framework is also a basis for performance monitoring services outlined by ITU Y.1731. Performance monitoring is used to validate Service Level Agreements based on frame loss, throughput, delay and delay variation.

Xelerated's Network Processors and programmable Ethernet switches can be programmed to support a range of service and performance OAM services. Delay measurement is of particular importance and is provided by Xelerated's Precision Time Solution. A highly accurate time stamping function on all incoming traffic brings high quality delay measurement.

The programmable pipeline and on-chip packet generators provide a pool of options for service OAM.

Session Learning

Xelerated's dataflow OAM processing capabilities are based on control program mechanism that can easily be leveraged in adjacent technology areas. One example is user sessions in broadband networks. They are enforced by service edge routers or access devices. The end-user's authentication traffic, based on DHCP or PPP, is forwarded to the authentication server, but when the traffic is forwarded, the user state should also be added to the corresponding filter table. This can be carried out in a similar manner as the MAC learning process explained above. 

As more and more trust is put on packet-based networks, carrier-class operations will have to depend on data plane operations that scale. The dataflow architecture can be equally leveraged for user data and for OAM and other control data processing.