Edge Switch Buffer Boxes for Fault Tolerant Stream Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In data center networks, recovering from a downstream server failure is time-consuming and bandwidth-intensive due to the need for upstream servers to retransmit data across the network to a new downstream server, affecting overall network performance and quality of service.
Innovation Solution
Implementing buffer boxes at the edge switch or aggregation switch level to locally buffer data streams, allowing new downstream nodes to access checkpoint data directly from these buffers rather than upstream nodes, reducing the recovery path and minimizing network traffic and bandwidth usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If upstream servers retransmit data across the network to a new downstream server after failure, then the new downstream server can recover the prior state, but recovery time and bandwidth consumption increase significantly
Solution Approach 1:
The patent implements preliminary action by having upstream servers continuously buffer data in network switches before it reaches downstream servers. This pre-positioning of data in intermediate network infrastructure allows the new downstream server to retrieve data from the nearest buffer rather than requiring retransmission across the entire network path, significantly reducing recovery time while maintaining fault tolerance
Solution Approach 2:
The patent introduces network switches as intermediary components that actively buffer data streams between upstream and downstream servers. These intermediaries serve as local data repositories that reduce the distance and time required for data retrieval during failure recovery, eliminating the need for long-distance retransmission while ensuring data availability
2Reliability
If upstream servers retransmit large amounts of data across the network after downstream server failure, then data recovery is achieved, but network bandwidth is significantly consumed
Solution Approach 1:
The patent applies local quality by distributing data buffers across multiple network switches at different locations in the network. When a downstream server fails, the system retrieves data from the nearest local buffer rather than retransmitting data across the entire network. This localized data access reduces network bandwidth consumption while maintaining complete data recovery capability
Solution Approach 2:
By pre-positioning data copies in network switches before failure occurs, the system eliminates the need for bandwidth-intensive retransmission during recovery. The preliminary buffering action ensures data is already at the retrieval location, converting a bandwidth-consuming retransmission operation into a simple local read operation
3Productivity
If data is buffered at edge switch or aggregation switch level, then recovery path is reduced and network traffic minimized, but additional buffer infrastructure is required
Solution Approach 1:
The patent makes network switches multi-functional by enabling them to perform both their traditional routing function and data buffering function. Rather than adding dedicated buffer infrastructure, the system utilizes the existing network switch infrastructure for dual purposes, reducing overall system complexity while achieving fast recovery speeds through local data access
Data Source
AI summary
A buffer box can be configured for an edge switch that is connected to one or more downstream nodes of a stateful stream processing system. Various upstream nodes can transmit data to downstream nodes for processing. At various locations in the data streams, the upstream nodes can include markers to indicate when a checkpoint should be generated or updated for the state of the data. Between markers, the data can be copied to a respective buffer box for the downstream node and edge switch. In the event the downstream node fails, the downstream node can obtain the checkpoint data to update state, and can receive the recent data from the local buffer box, rather than the corresponding upstream nodes. This can improve bandwidth usage and recovery time versus conventional recovery approaches, and can minimize impact of the recovery on other nodes across the network.