More Information...

Carrier Ethernet

Telecom Load Distribution

Application Note: Using FocalPoint in Telecom Backplane Fabrics (PDF)

White Paper: FM3000 Policy Engine (PDF)

White Paper: Ethernet Interconnects in ATCA Systems (PDF)

Using FocalPoint in a Telecom Backplane

Telecom backplanes lie within high bandwidth telecommunication systems and have traditionally used expensive proprietary switch fabrics due to specialized quality of service (QoS) and congestion management requirements. The FocalPoint family now offers telecom system designers the capability to design these fabrics using low-cost Ethernet switch silicon.

The FM3000 Series switch family provides advanced features such as congestion management, QoS and load distribution making it an ideal low cost solution for high bandwidth telecom backplane applications. The figure below illustrates a system using FocalPoint in a telecom backplanes including high-bandwidth switch cards and line cards. In this example, the FM3410 and the FM3212 are used as fabric interface devices on the line cards while the FM3224 is used as a backplane switch device.

The FM3410 on the 10G line cards route the ingress traffic from the NPU across four FM3224 devices. Since the FM3224 has 24 10G XAUI interfaces, this backplane can support up to 24 10G line cards. If four of these switch cards are used in the system, an FM3212 can route ingress traffic from two NPUs across eight of these devices allowing the fabric to support up to 24 20G line cards.

Fabric Routing

The NPU (Traffic Manager) queues incoming frames based on fabric destination and class of service in Virtual Output Queues. The frames are then routed through the fabric using either a standard Ethernet L2 header or using a Fulcrum proprietary header. Fabric traffic can be distributed per flow across multiple FM3224 switches. The Fulcrum switches use class-based output queuing in a shared memory implementation. The fabric can operate in packet mode or in cell mode if the NPU supports segmentation and re-assembly.

Fabric Scheduling

Each egress port in a FM3000 series device contains up to 8 class of service (CoS) queues which can support strict priority, minimum bandwidth guarantees using deficit weighted round robin and traffic shaping. Traffic classes can be assigned using various frame header fields.

Flow Control

Each egress port on the FM3224 can transmit the status of its class of service queues to all ingress NPUs using proprietary Virtual output queue Congestion Notification (VCN) frames. The FM3000 series devices also support link-level flow control using class-based pause frames. These mechanisms guarantee a lossless fabric without any head-of-line blocking.

Load Distribution

Multiple FM3224 switches can be used in the backplane in order to provide the required over-speed for the line cards. The ingress FM3000 devices can automatically distribute traffic across these switch devices using a highly efficient hash algorithm. In the event of switch chip or switch card failure, traffic can be automatically re-distributed around the failing device(s) providing N+1 redundancy. Since load distribution is flow-based, only flows to the failing switch will be affected.

Fabric Management

The FM3000 devices can be configured and controlled remotely using Fulcrum In-Band Management (FIBM) frames. In the figure above, one of the switch card CPUs can act as the active fabric manager while the other one acts as a standby fabric manager. The fabric manager is connected to the CPU interface of both FM3224 devices and can read or write to any FM3000 device register in the system using FIBM frames. Interrupts can also be received from any FM3000 device in the system using FIBM frames.

ATCA Backplane

The ATCA form factor has become popular among telecom system vendors due to the availability of standard blades for various telecom services. The figure below shows such a system using FocalPoint backplane devices. Since standard ATCA backplanes support two XAUI interfaces per line card, an NPU can be used to directly connect to the switch card through the backplane. Standard CPU blades can be used with PCI-express daughter cards connecting to the switch cards as shown. In some cases, NICs with dual XAUI interfaces can be used on these daughter cards. The standard Blade Center HT form factor can also be used in these applications with up to four switch cards per system.