Policy Engine for Dynamic Grid Resource Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional grid computing approaches are batch-oriented and lack the ability to dynamically adjust resources for scalable distributed persistent applications, leading to inefficiencies and inability to handle sudden increases in demand, such as during events like the 9/11 attacks.
Innovation Solution
Implementing a policy engine within a grid engine that dynamically allocates and deallocates physical and operating system resources based on performance and usage metrics, allowing for real-time scaling of resources for persistent applications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a larger amount of resources is statically allocated to the application to provide a large safety factor, then the application can handle increased load, but the excess resources would typically be idle most of the time, leading to waste and inefficiency
Solution Approach 1:
The patent implements dynamic resource allocation where the distributed resource manager continuously monitors application performance metrics (CPU usage, memory usage, response time) and automatically adjusts resource allocation in real-time. This allows resources to be allocated during high-demand periods and released during low-demand periods, resolving the contradiction between maintaining reliability for load surges and avoiding resource waste during normal operation.
Solution Approach 2:
The system employs feedback mechanisms where performance data from persistent applications is continuously collected and fed back to the distributed resource manager. Based on this feedback, the manager makes intelligent decisions about resource allocation, scaling up when performance degradation is detected and scaling down when resources are underutilized, thus balancing reliability and resource efficiency.
2Device complexity
If conventional batch-oriented approaches are used, then resource allocation is simplified, but the system cannot support users waiting with active sessions requiring results within a relatively short period of time
Solution Approach 1:
The patent transitions from static batch-oriented resource allocation to dynamic real-time allocation that responds to user session requirements. The distributed resource manager monitors session state and application performance, dynamically adjusting resource allocation to maintain responsive user interactions while preserving the underlying simplicity of the grid infrastructure.
Solution Approach 2:
The system creates a multi-functional resource management approach that can handle both traditional batch processing and interactive persistent applications. The same distributed resource manager infrastructure supports multiple workloads with different requirements, providing simplified resource allocation for batch jobs while enabling responsive user sessions through dynamic scaling when needed.
3Device complexity
If statically provisioned computing grids are used, then the distributed resource manager only needs to be apprised of workload and capabilities, but there is no way to determine based upon performance how the application is performing and whether additional resources need to be dedicated
Solution Approach 1:
The patent implements comprehensive feedback loops where the distributed resource manager continuously collects performance metrics from persistent applications including CPU utilization, memory consumption, response time, and throughput. This feedback enables the manager to detect performance degradation and automatically provision additional resources or scale existing ones, providing adaptability while maintaining manageable complexity through automated decision-making.
Solution Approach 2:
The system enables persistent applications to effectively self-manage their resource requirements through the feedback-driven automated provisioning process. The applications don't need to explicitly request resources; instead, the distributed resource manager autonomously monitors performance and allocates resources based on observed needs, reducing management overhead while improving adaptability.
Data Source
AI summary
There are provided mechanisms and methods for automating management of Scalable Distributed Persistent Applications (SDPA) in a grid computing environment. These mechanisms and methods make it possible for physical and operating system resources in the grid to be dynamically allocated and/or de-allocated based upon the results of monitoring performance and monitoring usage of physical and operating system resources.