Hierarchical Network
An architectural design that organizes a network into multiple tiers, such as access, aggregation and core layers. Its advantages lie in manageability and fault isolation, but its scalability is constrained by factors including oversubscription ratio, routing state growth and fault propagation.
Timeline (last 90 days)
Hierarchical networks in data center networking typically manifest as a three-tier architecture of access layer, aggregation layer, and core layer.
The advantages of hierarchical networks include manageability and fault isolation, because each layer interacts only with adjacent layers, so a single point of failure does not immediately affect the entire network.
In hierarchical networks, oversubscription ratios usually compound layer by layer, causing end-to-end available bandwidth to be far below the theoretical peak.
In hierarchical networks, traffic rerouting when a critical node fails may overload adjacent layers and create secondary congestion; the larger the scale, the harder such cascading effects are to predict.
Hierarchical networks often adopt a design of high bandwidth within racks and oversubscription between racks, but the global communication pattern of All-Reduce raises the share of cross-layer traffic, thereby amplifying the oversubscription bottleneck.
One approach to the scalability of hierarchical networks is to introduce more flexible routing or dynamic topologies so that traffic avoids congested layers; another is to reconsider the granularity of layering, for example replacing overly layered structures with flatter supernode designs.
Whether the scalability of hierarchical networks is limited by the topology itself or by the implementation of the control plane, the original HN post did not reach a conclusion.
Hierarchical networks trade locality for manageability, and their scalability depends on whether the balance point between the two can shift with scale.
All Facts (8)
Hierarchical networks in data center networking typically manifest as a three-tier architecture of access layer, aggregation layer, and core layer.
50%UnverifiedThe advantages of hierarchical networks include manageability and fault isolation, because each layer interacts only with adjacent layers, so a single point of failure does not immediately affect the entire network.
50%UnverifiedIn hierarchical networks, oversubscription ratios usually compound layer by layer, causing end-to-end available bandwidth to be far below the theoretical peak.
50%UnverifiedIn hierarchical networks, traffic rerouting when a critical node fails may overload adjacent layers and create secondary congestion; the larger the scale, the harder such cascading effects are to predict.
50%UnverifiedHierarchical networks often adopt a design of high bandwidth within racks and oversubscription between racks, but the global communication pattern of All-Reduce raises the share of cross-layer traffic, thereby amplifying the oversubscription bottleneck.
50%UnverifiedOne approach to the scalability of hierarchical networks is to introduce more flexible routing or dynamic topologies so that traffic avoids congested layers; another is to reconsider the granularity of layering, for example replacing overly layered structures with flatter supernode designs.
50%UnverifiedWhether the scalability of hierarchical networks is limited by the topology itself or by the implementation of the control plane, the original HN post did not reach a conclusion.
50%UnverifiedHierarchical networks trade locality for manageability, and their scalability depends on whether the balance point between the two can shift with scale.
50%