Random Next Iteration for Data Center Network Congestion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional approaches to managing network traffic in data centers and cloud computing environments often lead to network congestion due to synchronized communication patterns, resulting in increased latency and packet loss, especially in oversubscribed networks where large, expensive routers are underutilized most of the time.
Innovation Solution
The proposed solution involves dispersing workloads across multiple network switches to maximize buffering capacity and reduce congestion by using Random Next Iteration or Ordered Next Iteration techniques, which randomize or structure the communication order among hosts to minimize convoying behavior and incast events, thereby optimizing network performance with commodity switches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If large routers with significant buffer capacity are used to mitigate network congestion, then network reliability is improved, but device cost increases
Solution Approach 1:
The patent segments the buffering function from centralized large routers and distributes it across multiple commodity switches. Each switch maintains its own buffer, and the system leverages the aggregate buffering capacity of all switches together, eliminating the need for expensive centralized buffer capacity while maintaining network reliability.
Solution Approach 2:
The patent replaces expensive, high-capacity router buffers with multiple inexpensive commodity switch buffers. While individual switch buffers are smaller and less durable, their collective capacity matches or exceeds that of a single large router buffer, achieving the same reliability at lower cost.
2Productivity
If synchronized communication patterns are used among hosts, then coordination efficiency is improved, but network congestion increases
Solution Approach 1:
The patent implements periodic communication rounds where hosts take turns communicating with each other. In each round, a subset of hosts communicates while others remain quiet, creating periodic rather than continuous synchronized traffic patterns. This distributes network load over time, reducing peak congestion while maintaining coordination through the structured rotation of communication turns.
3Ease of manufacture
If commodity switches with small buffers are used instead of large routers, then device cost is reduced, but network reliability deteriorates
Solution Approach 1:
The patent merges the buffering resources of multiple commodity switches into a unified aggregate buffer system. By coordinating communication patterns across the network, the system effectively pools the small buffers of individual switches into a large virtual buffer capacity that matches or exceeds that of expensive dedicated router buffers, achieving high reliability with low-cost hardware.
Data Source
AI summary
Host machines and other devices performing synchronized operations can be dispersed across multiple racks in a data center to provide additional buffer capacity and to reduce the likelihood of congestion. The level of dispersion can depend on factors such as the level of oversubscription, as it can be undesirable in a highly connected network to push excessive host traffic into the aggregation fabric. As oversubscription levels increase, the amount of dispersion can be reduced and two or more host machines can be clustered on a given rack, or otherwise connected through the same edge switch. By clustering a portion of the machines, some of the host traffic can be redirected by the respective edge switch without entering the aggregation fabric. When provisioning hosts for a customer, application, or synchronized operation, for example, the levels of clustering and dispersion can be balanced to minimize the likelihood for congestion throughout the network.