Resource Quota Inheritance for Distributed Computing Latency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In distributed computing systems, managing computation resources efficiently to reduce latency and improve performance is challenging due to the overhead of resource requests and the limitations of commonly used software tools when handling large data sets, known as 'big data', which require optimization of resource utilization across node clusters.
Innovation Solution
Implementing resource quota inheritance and resource property inheritance mechanisms, where users can directly inherit resource quotas and properties from inactive users, reducing the need for network communication with the resource scheduler and enabling the reuse of resource products, thereby minimizing latency and enhancing application execution performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If resource requests are processed through the resource scheduler for each user, then resource allocation is centralized and controlled, but network communication overhead increases and latency increases
Solution Approach 1:
The system performs preliminary actions by pre-allocating resource quotas to users before they actually need resources. The resource controller maintains a quota pool with pre-computed resource allocations that can be instantly inherited by new users, eliminating the need for real-time resource scheduler communication during user startup.
Solution Approach 2:
The resource controller acts as an intermediary between users and the resource scheduler. It maintains a local quota pool that mediates resource allocation decisions, allowing users to inherit resources without direct communication with the resource scheduler, thus reducing network overhead and latency.
2Manufacturing precision
If resource quotas are allocated to each user individually, then resource management is precise, but the load on the resource scheduler increases
Solution Approach 1:
The resource scheduler performs preliminary resource quota calculations and allocations in advance, storing them in the resource controller's quota pool. This pre-computation reduces the scheduler's real-time processing load while maintaining precise resource allocation through the pre-calculated quotas.
Solution Approach 2:
The system uses copying by allowing new users to inherit resource quotas from existing users' quota pools rather than creating entirely new allocations. This copying mechanism reduces scheduler load while maintaining allocation precision through the inherited quota structures.
3Adaptability or versatility
If resource products are created fresh for each user, then resource allocation is flexible, but CPU and I/O costs increase
Solution Approach 1:
The system discards the need to create fresh resource products for each user by recovering and reusing resource products from the quota pool. Inactive users' resource products are preserved in the pool and can be recovered and reused by new users, reducing CPU and I/O costs while maintaining allocation flexibility.
Solution Approach 2:
The system uses copying by allowing new users to inherit and reuse resource products from the quota pool instead of creating entirely new resource products. This copying approach maintains allocation flexibility while significantly reducing the computational and I/O costs associated with creating fresh resources.
Data Source
AI summary
A resource request is sent from a user to a resource controller. The resource controller is resident on a node of a distributed computing platform comprising a plurality of nodes configured to collectively execute at least one application program. A quota pool associated with the resource controller is searched for a matching resource quota for the user to inherit. The resource quota is assigned to the user when a matching resource quota for the user to inherit is found in the quota pool. The resource request is forwarded to a resource scheduler that globally manages resources for the distributed computing platform, when a matching resource quota for the user to inherit is not found in the quota pool.