Optically Switched Network Distributed Arbitration Scalability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
As high-performance computing networks grow in size, existing central switch designs face challenges in scalability and thermal management, leading to increased complexity and thermal issues due to dense circuitry.
Innovation Solution
An optically switched network architecture with a virtual data plane providing any-to-all parallel connectivity and a virtual control plane organized as a ring, using a distributed-arbitration scheme where end-nodes independently manage network access through distinct wavelengths, minimizing contention and thermal challenges.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a central switch design is used to enable high-performance communication among computing nodes, then network performance is improved, but device complexity increases and scalability becomes difficult
Solution Approach 1:
The patent segments the centralized switch functionality into distributed components at each end-node. Each end-node independently performs arbitration and switching decisions, eliminating the need for a single complex central switch. This segmentation allows the network to scale by simply adding more end-nodes without increasing central switch complexity.
Solution Approach 2:
The patent extracts the arbitration and control logic from the central switch and places it at the end-nodes. The control plane is separated from the data plane, with end-nodes making independent arbitration decisions based on control tokens they receive, removing the complexity burden from the central switch fabric.
2Adaptability or versatility
If circuitry density is increased to accommodate more computing nodes, then network scalability is improved, but thermal management issues worsen due to hot spots
Solution Approach 1:
The patent segments the processing workload from a single dense central switch into distributed processing at multiple end-nodes. Each end-node handles only its local arbitration and switching decisions, spreading the thermal load across the network infrastructure rather than concentrating it in one location.
Solution Approach 2:
Each end-node independently manages its own arbitration and switching decisions without requiring intensive processing at a central location. This self-service approach distributes the computational burden and associated heat generation across all end-nodes, improving thermal management.
3Device complexity
If a distributed arbitration scheme is used to reduce central switch complexity, then device complexity is reduced, but control overhead increases due to token circulation
Solution Approach 1:
The patent merges the control plane and data plane into a unified optical infrastructure. The same optical network that carries data also carries control tokens, eliminating the need for separate control wiring and reducing overall control overhead. Control tokens are embedded in the optical signal stream and processed at end-nodes without requiring additional physical channels.
Data Source
AI summary
We disclose a method for controlling access to an optically switched network, which connects N end-nodes, and is organized into a virtual data plane and a virtual control plane, which both communicate through the same underlying physical optical network. The virtual data plane provides any-to-all parallel connectivity for data transmissions among the N end-nodes, and the virtual control plane is organized as a ring that serially connects the N end-nodes, wherein a control token circulates around the ring. During operation, an end-node in the ring receives the control token, which includes a destination-busy vector with a busy flag for each of the N end-nodes. If the end-node has data to send and the busy flag for the destination end-node is not set, the system: sets the busy flag; commences sending the data to the destination end-node; and forwards the control token to a next end-node in the ring.