BCube Server-Centric Network Architecture for Modular Data Centers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modular data centers face challenges with bandwidth-intensive requirements and server/switch failures, necessitating a network architecture that supports high inter-server bandwidth, aggregate throughput, and graceful performance degradation.
Innovation Solution
A server-centric network architecture, known as the BCube network, where each server acts as both an end host and intermediate relay node with multiple ports, utilizing a hybrid Butterfly Cube structure to provide high aggregate network capacity, fault tolerance, and load balancing through multiple parallel paths and a fault-tolerant source routing protocol.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a typical two-level tree structure is used to interconnect servers in an MDC, then the network architecture is simple to implement, but the inter-server bandwidth and aggregate throughput are insufficient
Solution Approach 1:
The network architecture is segmented into multiple levels (level-0, level-1, level-2) with different functional roles. Level-0 connects servers to local switches, level-1 provides intermediate routing, and level-2 provides core connectivity. This segmentation allows high bandwidth through multiple parallel paths while keeping each segment's complexity manageable
Solution Approach 2:
The patent transitions from a two-level tree structure to a three-level BCube architecture, adding an intermediate level-1 network. This dimensional change creates multiple routing dimensions and parallel paths between servers, significantly increasing aggregate throughput while distributing complexity across multiple manageable levels
2Reliability
If switches are used to interconnect all servers in an MDC, then connectivity is provided, but the system becomes vulnerable to single points of failure and lacks fault tolerance
Solution Approach 1:
Different parts of the network have different functional qualities optimized for their roles. Level-0 switches handle local server connectivity, level-1 switches provide intermediate routing with fault isolation, and level-2 switches provide core redundancy. This local optimization allows fault tolerance without requiring every component to be equally complex
Solution Approach 2:
The architecture pre-establishes multiple parallel paths and redundant connections before failures occur. When a switch or server fails, traffic is automatically rerouted through alternative paths that were already in place, providing graceful degradation without requiring reactive complexity
3Productivity
If servers have multiple network ports and act as relay nodes, then high inter-server throughput is achieved, but the routing complexity increases
Solution Approach 1:
Routing tables are pre-computed and cached before traffic flows occur. The BCube routing protocol establishes forwarding paths in advance based on the known network topology, allowing servers to make simple table-lookup forwarding decisions rather than computing routes in real-time. This preliminary action enables high throughput while keeping per-packet complexity low
Solution Approach 2:
The patent introduces a dedicated routing protocol (BCube routing) as an intermediary layer between the physical network topology and data traffic. This protocol abstracts the complexity of multi-path routing by providing simple source-routing rules and pre-computed forwarding tables, allowing servers with multiple ports to achieve high throughput without complex real-time routing logic
Data Source
AI summary
Disclosed are systems and methods for network architecture that is a server-centric network architectural design.