8.9 Evaluating the Single Control Card Multiple Line Card Architecture

8.9 Evaluating the Single Control Card + Multiple Line Card Architecture

There are several advantages to the single control card + multiple line card architecture, including a single point of control for the control plane and management protocols, distribution of information like routing table information, relatively easy software migration from a single-processor architecture (if we use messages), and so on.

However, the architecture also has some disadvantages:

1. The control card is a single point of failure—if it goes down, so do all the control plane and management plane protocols.

2. The control card processor can become overloaded due to the large number of control plane tasks that need to be implemented. Some protocols such as OSPF and IS– IS require complex SPF (Shortest Path First) calculations, potentially tying up the control processor for long periods.

3. The control card–to–line card control interface is a potential bottleneck, since the line cards need to forward messages destined to the router on this interface to the control card.

One way to address these issues is to add additional control cards. Some systems follow such a scheme for scalability—for example, one control card could run all the interior routing protocols (OSPF, IS–IS and RIP) while another could run the exterior routing protocol like BGP and also the multicast routing protocols like PIM (Protocol Independent Multicast), MOSPF (Multicast Open Shortest Path First), and DVMRP (Distance Vector Multicast Routing Protocol). The line cards are aware of this division and send the received packets to the appropriate control card. Note that the protocols continue to have a complete system view since they communicate with all the line cards.

The power provided by the additional control cards comes with its price—the need for two control cards to exchange information adds complexity. The RTM needs to be common, since it will be used to determine the policy for the routes and route redistribution and pass this information to the IP Switching Task on the individual line cards. With the routing tasks split across multiple control cards, this becomes a challenge. Another issue to consider is the distribution of management function between the control cards.

Despite the complexity, this solution is more scalable. Hence, some designers use a completely distributed architecture for their multi-processor system, as discussed next.