Front-End Switch Load Balancing via IP Reassignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current load balancing techniques require hardware load balancer devices and result in connection resets when compute nodes fail or return online, leading to inefficiencies and increased public IP address requirements.
Innovation Solution
Implementing a method that uses a front-end switch to route requests to internal IP addresses assigned to compute nodes, with reassignment and reclaiming of these addresses based on compute node status, maintaining connections and balancing workload without relying on hardware balancers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If hardware load balancer devices are used to distribute requests, then load balancing is achieved, but device complexity and cost increase
Solution Approach 1:
The patent extracts the load balancing function from dedicated hardware devices and relocates it to standard network switches. The switch uses a hash table to map public IP addresses to internal IP addresses of compute nodes, eliminating the need for specialized load balancer hardware while maintaining distribution capabilities.
Solution Approach 2:
The patent introduces an Address Resolution Protocol (ARP) table as an intermediary mechanism between network switches and compute nodes. This ARP table enables dynamic mapping and reassignment of IP addresses without requiring hardware changes, facilitating seamless load balancing and failover.
2Reliability
If compute nodes fail or return online, then system resilience is improved, but connections are reset leading to service disruption
Solution Approach 1:
The system performs preliminary actions by pre-establishing the ARP table with mappings between public IP addresses and internal IP addresses before failures occur. When compute nodes fail or return, the pre-configured ARP table enables automatic reassignment without resetting connections, maintaining service continuity.
Solution Approach 2:
The patent implements a mechanism where internal IP addresses are temporarily discarded from failed compute nodes and recovered/reassigned to available compute nodes through the ARP table. This allows the system to recover from failures while maintaining connection integrity, as the mapping layer abstracts the physical node changes.
3Adaptability or versatility
If public IP addresses are assigned to each compute node for direct access, then node accessibility is improved, but the number of required public IP addresses increases
Solution Approach 1:
The patent merges multiple internal IP addresses under a single public IP address using a hash table mapping mechanism. The front-end switch receives requests for the public IP address and distributes them to appropriate internal IP addresses based on the hash function, enabling one public IP to serve multiple compute nodes simultaneously.
Solution Approach 2:
The public IP address becomes a universal entry point that can route to any compute node in the cluster. The ARP table and hash table mechanisms enable this single public IP to function as a multi-functional gateway, directing traffic to whichever internal node is appropriate based on current system state and load conditions.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for receiving a plurality of requests from a plurality of client devices sent to a public internet protocol address; routing the requests to one or more internal Internet protocol addresses, where the one or more internal Internet protocol addresses correspond to one or more compute nodes, and where the one or more compute nodes process the routed requests; determining a particular compute node is offline, where the particular compute node is assigned one or more internal Internet protocol addresses; in response to the determination that the particular node is offline, reassigning the one or more internal Internet protocol addresses corresponding to the particular compute node to one or more different compute nodes, where requests communicated to the reassigned internal Internet protocol addresses are communicated to the one or more different compute nodes for processing.