There are three main shortcomings with a pure non PBT/PBB enabled Ethernet MAN approach:By design, layer 2 switches use fixed tables to direct traffic based on the MAC address of the endpoints. As the network gets larger, the number of MAC address transiting through the network may grow beyond the capacity of the core switches. If the core table gets full, the result is a catastrophic loss of performance due to the flooding of packets over the entire network structure.
This can be overcome to some degree by smart network design and keeping your network segments and rings small enough to support the MAC table limitations of the equipment. In a pure ethernet network, the network should be designed in a modular grouping where your less expensive, smaller MAC table devices are in small geographically significant segments connected by larger aggregation devices which are interconnected that support two tag manipulation and very large MAC tables. This design keeps locally geographically significant segments interconnected with less expensive equipment, and larger geographically connected areas interconnected with more expensive, more feature laden equipment. This keeps the MAC tables small and helps keep the pure ethernet network scalable.
Network stability is relatively fragile, especially if compared to the more advanced SDH and MPLS approaches. The recovery time for the standard spanning tree protocol is in the range of tens of seconds, much higher than what can be obtained in the alternative networks (usually a fraction of second). There are a number of optimizations, some standardized through the IEEE, and others vendor-specific, that seek to alleviate this problem. The clever use of such features allow the network to achieve good stability and resilience, at the cost of a more complex configuration and possible use of non-standard, vendor-specific, mechanisms. Some vendor's implementations of RSTP achieve sub 50ms convergence in typical sized rings. RSTP also provides for easy deployment of complex designs such as multi-ring, figure eight, etc. If designed appropriately, in many networks the fragility in this network design can be overcome without the additional expense of MPLS.
Traffic engineering is very limited. There are few tools to manage the topology of the network; also, the fact that forwarding is done hop-by-hop, added to the possibility of broadcasts even for unicast packets (for instance, while learning new addresses), makes predicting the real traffic pattern very difficult for a networking novice. Custom tools, such as topology maps that outline where blocking ports occur in the network during normal and backup conditions may need to be built to fully understand and troubleshoot the network quickly.
December 16, 2011