 |
|
|
Cariden's technology incorporates state-of-the-art optimization methods, designed from
the ground up to meet the complex specifications of modern network design, engineering and control.
Specific areas of Cariden's technological leadership include:
CONTROL OF PURE IP NETWORKS: from Art to Science
In pure IP networks, a few simple levers, the IGP and BGP metrics and preferences, are used to control link utilizations and avoid congestion.
Between this simple input and output is a vast black box in which huge numbers of traffic routings interact in complex ways.
Cariden's technology makes the black box transparent:
|
 |
Demand matrices are modeled and predicted using observed utilizations and behavior of the network.
|
|
The network is simulated under SRLG failure scenarios, service class interactions are modeled to ensure QoS, bottlenecks are identified.
|
|
Metrics are optimized to achieve desired engineering goals: resilient throughput maximization, bounds on latencies, desired service class mix.
Performance typically approximates that achievable by complete explicit routing control.
|
|
With the transparency, predictability and optimization provided by the Cariden system, the art of pure IP network control has become a science.
OPTIMIZING EXPLICT ROUTING AND PROTECTION
In explicitly routed network, a large (n2) number of individual routings must be maintained, and their interactions predicted and controlled.
Cariden provides the visual and computational aid to make this process manageable.
Cariden's inbuilt explicit routing optimizations are fast, and achieve more than 95% the efficiency of the best possible (multicommodity flow) routings in typical networks.
Static (hotstandby and fast-reroute) protection methods are optimized to maximum efficiency by Cariden's technologies.
Cariden also provides for mixing explicit and IP routing to provide robust hybrid routing solutions.
OVERVIEW OF MATHEMATICAL ADVANCES
Traditional network optimization methods such as dynamic and mixed-integer programming work well for small network optimizations.
Large modern networks with complex protection and stability requirements need more.
MATE uses highly efficient approximation methods to control the explosion of complexity and lack of predictability normally associated
with the solution of large-scale network optimization problems.
These methods include probabilistic approximations, convex relaxations, and heuristic methods developed through experimentation
on a range of planned and currently operational networks.
MATE uses statistical prediction methods to model future network demand and usage.
Robust convex optimizations incorporate both these predictions and their estimated uncertainty in their design and engineering recommendations.
|
|
