Generic Partitioning Manager for Data Center Resource Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing lease management systems in data centers face issues with resource partitioning across nodes, where grant messages can be interrupted and lead to confusion over resource ownership, especially due to multiple instances of logical resources and load imbalances.
Innovation Solution
A generic partitioning manager that interfaces with owner nodes through an owner library and lookup nodes through a lookup library, using leases to allocate resources based on load measurements and node status, ensuring efficient resource distribution across multiple owner nodes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple instances of logical resources are created for partitioning, then resource distribution flexibility is improved, but confusion over resource contents and ownership accuracy deteriorates
Solution Approach 1:
The system segments logical resources into distinct lease instances, where each lease represents a specific resource partition. This segmentation allows multiple instances to coexist while maintaining clear boundaries through lease identifiers and ownership tracking mechanisms, preventing confusion about resource contents.
Solution Approach 2:
The partitioning manager implements feedback mechanisms through lease renewal processes and ownership verification protocols. Lease holders must actively renew their leases and the system provides feedback on ownership status, ensuring accurate tracking of resource contents across multiple instances.
2Productivity
If grant messages are used for resource allocation, then resource assignment capability is improved, but message interruption and loss occur
Solution Approach 1:
The system implements beforehand cushioning by requiring lease holders to actively renew their leases before expiration. This proactive renewal process ensures continuous ownership tracking and prevents the system from assuming resource unavailability due to potential message losses, cushioning against reliability issues.
Solution Approach 2:
The lease management system operates with a degree of self-service through automatic lease expiration handling and proactive renewal requirements. The system automatically tracks lease status and can respond to expiration events without requiring constant external intervention, improving reliability of resource allocation.
3Adaptability or versatility
If resources are partitioned across multiple owner nodes, then system scalability is improved, but load imbalance and partitioning complexity increase
Solution Approach 1:
The partitioning manager implements universality by handling multiple partitioning scenarios through a single unified lease management framework. The same lease-based mechanisms work for various partitioning strategies (static, dynamic, load-based), reducing the complexity of managing different partitioning approaches across scalable systems.
Data Source
AI summary
The application discloses a generic partitioning manager for partitioning resources across one or more owner nodes. In illustrated embodiments described, the partitioning manager interfaces with the one or more owner nodes through an owner library. A lookup node or application interfaces with the partitioning manager through the lookup library to lookup address or locations of the partitioned resources. In illustrated embodiments, resources are partitioned via the partitioning manager in response to lease request messages from an owner library. In illustrated embodiments, the lease grant message includes a complete list of the leases for the owner node.