Because of the virtual nature of the star the need for superprocessor at the core is unclear. It becomes clearer under application. If Metro Fusion serves business data needs only, then the state space of the entire network need not be known. If Metro Fusion attempts to scale up to multimedia across ISP cable broadband networks, the bit volume is so great that massive caching at the headend becomes critical. The superprocessor needs then migrate to the edge.
Consider two cities connected by a sixteen lane freeway where the law does not permit lane changing on the freeway. When traffic arrives at a city and enters, it is subjected to traffic lights. Local traffic is uselessly redundant. You have people in cars taking their children to school because someone once hijacked a school bus. Some of the traffic from the remote city is bound for the school too, but they only use the one allotted lane briefly. They either return or they park their cars and do something else while Johnny doesn't learn to add.
This school is growing in popularity. The freeway lane coming into the city is increasingly backed-up waiting for a parking space and increasingly in a wait state to get back onto the freeway home or to the next city. If the Core manages the rate of freeway access, the lane back-up will be reduced, but that requires knowing the wait state parameters governing traffic at the node. If the freeway lane is egress/ingress net flux independent, because the local traffic is managed through cached loop redundancy, lane traffic can continue to grow uninterrupted. The Core does not need to be involved with stress management at the focal point, the headend. It is still true though that as traffic increases linearly per fiber, the amount of management at each node rises at (k*p)^n where n = # of lanes, k = normalized lane volume detected at node, p = normalized redundant plus local origin traffic. It seems there is decision making critically necessary at the headend node in order to prevent traffic management there from blowing up from a scale burst. The management would take the form of price-speed-priority QoS and would require superprocessor for real time implementation.
A synergy of scale may exist between SilkRoad and DD over the non-metro WAN. Extend DD globally. Implement SR on the long haul using the SR optical pump off loading to DD Metro Fusion at NAP. One fiber hauls ATT's entire worldwide broadband network and another handles all of SBC's. Separate but equal. |
