Data Center Interconnect Switch Latency Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional data center interconnect (DCI) systems face challenges in maintaining low buffer occupancies globally, which can severely impact delay-sensitive traffic flows due to queuing behind long flows, and existing solutions are application-specific, require hardware modifications, and do not consider global constraints across flows and devices.
Innovation Solution
A method is introduced to dynamically resize line-side capacity in DCI networks by calculating an undersubscription factor based on client-side and line-side capacity values, using bandwidth resizing devices to optimize bandwidth allocation across all connected switches, thereby reducing latency and buffering while maintaining throughput.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If packet buffering is increased in DCI to maintain throughput, then network bandwidth is preserved, but latency for delay-sensitive short flows increases severely due to queuing behind long flows
Solution Approach 1:
The invention segments the network bandwidth into multiple wavelength channels (lambda channels) that can be independently allocated and managed. By dividing the bandwidth resource into discrete segments, the system can allocate specific wavelength channels to different traffic types (delay-sensitive vs. throughput-sensitive flows), preventing short flows from being blocked by long flows in a shared buffer while maintaining overall network throughput.
2Productivity
If conventional buffer management is used to maintain global throughput, then network bandwidth utilization is optimized, but buffer occupancies cannot be kept low, causing latency issues
Solution Approach 1:
The invention applies different buffer management qualities to different wavelength channels or traffic classes. Instead of uniform buffer management, the system can maintain low buffer occupancies for wavelength channels carrying delay-sensitive traffic while allowing higher buffer occupancies for channels carrying throughput-sensitive background traffic, thus achieving both low latency and high bandwidth utilization simultaneously.
3Loss of time
If existing latency reduction techniques are applied at one data center, then local latency is improved, but buffering increases in other connected data centers
Solution Approach 1:
The invention implements a centralized or coordinated buffer management system that monitors buffer occupancy levels across multiple data centers and dynamically adjusts wavelength channel allocations accordingly. When one data center experiences low latency conditions, the system receives feedback and redistributes buffer resources to other connected data centers, preventing excessive buffering elsewhere while maintaining overall network performance.
Data Source
AI summary
Systems, methods, and devices for managing latency in a network with a plurality of switches, each switch having client side ports and line side ports. A required bandwidth for each link between connected pairs of the plurality of switches is received. A client-side capacity value for each switch is received. An initial undersubscription factor is calculated based on the required bandwidths and the client-side capacity values. A desired undersubscription factor is calculated for each switch based on the initial undersubscription factor and the client side capacity values. A desired bandwidth is calculated for each link between connected pairs of the plurality of switches based on the required bandwidths and the desired undersubscription factors.