Load Balancer VM Stress Notification via Measurement Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional load balancers in Software-Defined Networking (SDN) lack visibility into the state of virtual machines (VMs), leading to inefficient distribution of network requests, as they do not consider the current load and resource utilization of VMs when balancing traffic.
Innovation Solution
A method is introduced to collect real-time measurement and statistical data from computing resources of a data compute node (DCN), allowing for the identification of stress levels and notification to a central monitoring agent, which then adjusts load balancing and security scans to avoid overloading VMs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional load balancers distribute requests based solely on network traffic, then load balancing is simple to implement, but the load balancing efficiency deteriorates because VM state visibility is lacking
Solution Approach 1:
The patent introduces a measurement data monitor as an intermediary component that collects measurement data from multiple sources (VM resource usage, host machine status, network traffic) and provides a unified view to the load balancer. This intermediary layer enables informed load balancing decisions without requiring the load balancer to directly manage complex data collection from multiple sources, thus improving load balancing efficiency while managing system complexity.
Solution Approach 2:
The measurement data monitor serves multiple functions: it collects VM resource usage data, host machine status, and network traffic information, then consolidates this data for use by both load balancers and security scanners. This multi-functional component improves overall system efficiency by eliminating redundant data collection efforts and providing a universal data source for multiple applications.
2Reliability
If security scans are performed on all DCNs regardless of stress level, then security coverage is comprehensive, but system reliability deteriorates due to additional load on already stressed resources
Solution Approach 1:
The system performs preliminary assessment of DCN stress levels using measurement data collected from resource usage metrics before initiating security scans. By evaluating the current state of computing resources in advance, the system can determine whether a DCN can withstand the additional load of a security scan, thereby preventing scans on already stressed resources and maintaining system reliability.
Solution Approach 2:
The security scanning policy becomes dynamic based on real-time measurement data. Instead of a static approach where all DCNs are scanned uniformly, the system adapts the scanning behavior according to the current stress level of each DCN, adjusting the scanning schedule and intensity to match the available resources and minimize harmful effects on system reliability.
3Productivity
If load balancers lack visibility into VM state, then the load balancing mechanism is simple, but resource utilization deteriorates as overloaded VMs continue to receive traffic
Solution Approach 1:
The patent implements a feedback mechanism where measurement data about VM resource usage (CPU, memory, disk I/O) is continuously collected and fed back to the load balancer. This feedback loop enables the load balancer to make informed decisions about traffic distribution, directing new connections away from overloaded VMs and toward underutilized resources, thereby optimizing overall resource utilization while maintaining simple load balancing logic.
Data Source
AI summary
Some embodiments provide a method for an end machine, that implements a distributed application, to redirect new network connection requests to other end machines that also implement the distributed application. The method receives a set of measurement data from a set of resources of the end machine and determines whether a measurement data received from a particular resource has exceeded a threshold. When the measurement data has exceeded the threshold, the method notifies a load balancer that balances new requests for connection to the distributed application between the end machines. The notification causes the load balancer not to send any new connection request to the end machine and redirect them to other end machines.