Multi-chassis core routers provide customers with the following operational advantages:

1. Investment protection on existing core network equipment.

2. Minimal disruption to existing core networks while upgrading.

3. Extreme bandwidth scalability.

While competitive solutions offer similar advantages, only Juniper’s T-series multi-chassis architecture provides enhanced fault isolation by combining all stages of a 3-stage Clos network inside a single Fabric Switch Chassis.



Multi-chassis core systems vary from single chassis versions primarily by their use of the following technologies:

1. 3-stage non-blocking Clos networks. While a single chassis system has only a single stage fabric that the network data transverses, multi-chassis systems use three. The use of three fabric stages provides the large number of endpoint packet forwarding engines (PFEs) needed for large scale routers.

2. Short reach, high density fabric fiber interconnects. This interconnect technology varies from standard WAN/LAN optics by providing a very large amount of data transfer at a low cost. Typically, multi-fiber cables are used for this technology incorporating arrays of VCSEL to achieve the requirement of low cost, short reach transmission. In addition, since these are only used to connect the fabric, the transmission protocols can be proprietary.



A typical multi-chassis router will require hundreds of these optical fabric interconnects. Therefore, management and fault tolerance of these cables are critical for the reliability of these systems. Where these cables sit in the three stages of the Clos network are key to how faults on these cables are propagated through the system.



All competitive multi-chassis core routing solutions place the F1 and F3 stages of the Clos network in the Line Card Chassis node. Then, the F2 stage alone sits in the center Fabric Switch Chassis. The problem with this approach is that because the fabric optical interconnects sit in-between the F1/F2 and F2/F3 stages, any fault with these cables “spray” the errors across multiple packet forwarding engines (PFEs) . A complete failure of a cable will also lead to bandwidth reduction sprayed across multiple PFEs.



Figure 1: Competitive system packaging of F1/F2/F3 fabric stages

Juniper takes a different approach.



Juniper’s architecture places all 3 stages of the Clos network inside the Fabric Switch Chassis. With this system, a fault or failure on a fabric interconnect cable will only manifest on a single PFE. The errors on one of these cables do not get “Sprayed” across the F1->F2 or F2->F3 stages. They stay contained to a single PFE.





Figure 2: Juniper’s T-series system packaging of F1/F2/F3 fabric stages



It is important to note that the total fabric bandwidth needed to transverse the inter-chassis optical cables remains the same in either approach. No matter if the F1 and F3 stages are in the LCC or FSC, the number of optical cables remains the same.



In summary, Juniper’s multi-chassis core router architecture provides a higher level of fault isolation due to the fact that it contains all 3 stages of a Clos network inside the Fabric Switch Chassis.

References:

Background on Clos networks: http://en.wikipedia.org/wiki/Clos_network