Directed Graph Resource Manager for Cloud Fault Tolerance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cloud computing systems face challenges in managing and recovering from resource failures and hardware upgrades, which can disrupt long-running operations and datacenter failures, requiring improved fault tolerance and scalability.
Innovation Solution
A resource management system that represents distributed computing resources as nodes in a directed graph, allowing for message passing to allocate or deallocate resources, and includes a message handler task to process messages persistently, enabling migration of resources and improving fault tolerance and scalability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If cloud computing systems use traditional resource management approaches, then system operations can be maintained, but fault tolerance is insufficient and long-running operations cannot recover from resource failures
Solution Approach 1:
The system segments resources into distinct types (compute resources, storage resources, network resources) and represents them as separate nodes in a directed graph. This segmentation allows independent tracking and management of each resource type, enabling the system to isolate failures and maintain operation continuity without managing all resources as a monolithic complex system.
Solution Approach 2:
The patent introduces a resource manager as an intermediary component that mediates between resource allocation decisions and actual resource allocation. This mediator tracks resource states in a directed graph and manages message passing between different resource types, simplifying the complexity of coordinating failures across multiple resource types while improving fault tolerance.
2Reliability
If the system implements comprehensive resource tracking and failure recovery mechanisms, then fault tolerance improves, but scalability is limited due to increased system complexity
Solution Approach 1:
The resource manager implements a universal approach by representing all resource types (compute, storage, network) as nodes in a single directed graph structure. This multi-functional representation allows the same graph-based mechanism to track and manage diverse resource types uniformly, enabling scalability without requiring separate tracking systems for each resource type while maintaining comprehensive fault tolerance.
Solution Approach 2:
The patent transitions from traditional flat resource management to a hierarchical directed graph structure where resources are organized in parent-child relationships across multiple dimensions. This dimensional organization allows the system to scale by adding resources at different levels of the graph without increasing the complexity of tracking and management, as the hierarchical structure provides natural organization and isolation.
3Ease of manufacture
If the system uses rigid resource management architecture, then implementation is simpler, but flexibility to define new message and resource types is reduced
Solution Approach 1:
The resource management architecture is designed as a dynamic system where the directed graph structure can adapt to accommodate new resource types without requiring changes to the core architecture. The graph can dynamically add nodes and edges to represent new resource types, and the message passing system can accommodate new message formats while maintaining the same fundamental allocation and deallocation mechanisms, thus preserving implementation simplicity while enabling flexibility.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for managing resources in a computing system. A plurality of distributed computing system resources are represented as nodes in a graph. Links between the nodes are defined, where each link in the links representing ownership or usage by a parent node of a respective resource represented by a child node. A message is sent from a source node in the graph to a target node in the graph wherein the source node is linked directly or indirectly to the target node. The message is processed by a message handler task associated with the target node, wherein processing includes allocating or deallocating a respective resource represented by the target node.