Exponential Weighted Round Robin Resource Manager
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computing systems do not effectively manage resource utilization across multiple resources, leading to uneven load distribution and inefficient resource selection, as they fail to consider the health and load-handling capabilities of resources beyond latency metrics.
Innovation Solution
A resource manager system that receives feedback on resource loads, calculates weights using an exponential weighted round robin method, and selects resources to evenly distribute load, ensuring fair usage and dynamic control of resource lifetimes and availability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If resources are selected without considering load metrics, then resource selection is simple, but load distribution becomes uneven and response time increases
Solution Approach 1:
The system implements feedback by continuously monitoring load metrics from resources and using this information to dynamically adjust selection probabilities. The load metric feedback loop enables the resource manager to adapt to changing resource states, ensuring balanced load distribution while maintaining responsive resource selection.
Solution Approach 2:
The resource selection mechanism transitions from static to dynamic by adjusting selection probabilities based on real-time load metrics. The probability of selecting a resource changes dynamically according to its current load state, allowing the system to adapt to varying resource capacities and maintain optimal performance.
2Productivity
If load-based weight calculation is implemented, then resource utilization becomes even, but calculation complexity increases
Solution Approach 1:
The system changes the selection parameter from uniform probability to load-based probability. By transforming the selection criterion into a probability model where P(selection) ∝ (1 - load_metric), the system achieves balanced resource utilization while maintaining a mathematically tractable calculation approach.
3Productivity
If exponential weighted round robin is used, then load distribution is optimized, but selection computation time increases
Solution Approach 1:
The system transforms the resource selection problem into a probability calculation problem based on load metrics. By using the probability formula P(i) = (1 - load_i) / Σ(1 - load_j), the system achieves optimal load distribution while keeping computations relatively simple and scalable to multiple resources.
Data Source
AI summary
A system includes a processor and memory to execute an application. The application receives feedback from a target regarding ability of a plurality of resources of the target to service requests from one or more clients. The feedback includes a metric indicative of a load of each of the resources. The application calculates weights for the resources based on the feedback. A weight for a resource is based on a product of a first term that determines a maximum difference in probabilities of selection between two resources and a second term including an exponent that is a difference between a current load of the resource and a current minimum load across the resources determined based on the feedback. The application selects, for servicing a request from one of the clients, one of the resources in round robin manner based on the weights to evenly utilize the plurality of resources.