Cluster Resource Sharing Control for Fair Multi-Tenant Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In multi-tenant computer clusters, the distribution of computing resources across customers is traditionally unequal and uncontrollable due to unpredictable customer requests, leading to unfair allocation based on the FIFO method, where customers with more requests receive a greater share of resources.
Innovation Solution
Implementing a flow-control instance that manages and controls the sharing of consumable resources through a balanced-utilization partitioning scheme, which dynamically allocates resources based on empirical analysis and priority queues, ensuring equitable distribution among active customers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If FIFO method is used to service requests, then requests are serviced in order of arrival, but customers with more requests receive a greater share of computing resources resulting in unequal distribution
Solution Approach 1:
The patent implements dynamic resource allocation by continuously monitoring and adjusting resource distribution based on real-time request patterns and customer needs. The system transitions from static FIFO processing to dynamic scheduling that adapts to changing conditions, ensuring fair resource sharing while maintaining operational simplicity.
Solution Approach 2:
The system incorporates feedback mechanisms that monitor resource usage patterns, request queues, and customer needs to continuously adjust allocation decisions. This feedback loop enables the system to correct unequal distributions and maintain fairness while preserving the simplicity of request processing.
2Device complexity
If requests are serviced sequentially in FIFO manner, then queue management is simple, but the distribution of computing resources across customers is not controllable
Solution Approach 1:
The patent segments the resource allocation process into distinct controllable components, including priority queues, resource pools, and allocation policies. This segmentation maintains the simplicity of queue management while introducing granular control over resource distribution through multiple independent management layers.
Solution Approach 2:
The system introduces intermediary control mechanisms between request queuing and resource allocation, such as priority schedulers and resource managers. These intermediaries maintain the simplicity of FIFO processing while enabling controllable and adaptable resource distribution through additional management layers.
3Manufacturing precision
If empirical analysis and priority queues are implemented for resource allocation, then fair and controlled resource distribution is achieved, but system complexity increases
Solution Approach 1:
The patent designs the flow control system with multi-functional components that handle multiple tasks including empirical analysis, priority queue management, and resource allocation. This universality reduces overall system complexity by consolidating functions into integrated components while maintaining fair and controlled resource distribution.
Solution Approach 2:
The system manages complexity by dynamically adjusting allocation parameters and priorities based on empirical analysis results. Rather than implementing complex fixed structures, the system uses parameter changes and adaptive policies to achieve fair resource distribution, reducing structural complexity while maintaining allocation fairness.
Data Source
AI summary
In one embodiment, a method includes empirically analyzing a set of active reservations and a current set of consumable resources belonging to a class of consumable resources. Each active reservation is of a managed task type and includes a group of one or more tasks requiring access to a consumable resource of the class. The method further includes, based on the empirically analyzing, clocking the set of active reservations each clocking cycle. In addition, the method includes, responsive to the clocking, sorting a priority queue of the set of active reservations.