Data Center Server Resource Allocation via Auction Agents
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The complexity of pairing servers with services in large-scale data centers leads to inefficient resource allocation, resulting in unused resources and increased operating costs due to varying service requirements and the need for frequent server additions.
Innovation Solution
A method and apparatus for self-organization of data centers, where service agents and server agents dynamically associate server resources with services using a probabilistic transition dynamic, allowing for efficient resource allocation through a Markov chain-based auction process, ensuring proportional fairness in resource distribution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If servers are manually paired with services in large-scale data centers, then resource allocation can be controlled, but the complexity of pairing increases and resource allocation efficiency decreases
Solution Approach 1:
The patent implements self-service through autonomous agents that automatically perform server-service pairing without human intervention. Service agents bid for server resources and server agents allocate resources based on bids, creating a self-organizing system that reduces operational complexity while improving allocation efficiency through automated market-based mechanisms.
Solution Approach 2:
The patent introduces an intermediary auction mechanism that mediates between service agents and server agents. This market-based intermediary translates service resource requirements into bids and server resource allocations into prices, enabling complex multi-server pairings to be resolved through simplified bid-price interactions rather than direct manual configuration.
2Adaptability or versatility
If service requirements vary over time and new services are added, then system adaptability improves, but resource allocation complexity and operating costs increase
Solution Approach 1:
The patent implements dynamics by making the server-service pairing configuration changeable over time through continuous bidding and allocation. Service agents can adjust their bids based on changing requirements, and server agents can reallocate resources dynamically, allowing the system to adapt to new services and varying demands without manual reconfiguration complexity.
Solution Approach 2:
The patent employs feedback mechanisms where server agents provide allocation status and pricing information back to service agents, enabling them to adjust their bids accordingly. This closed-loop feedback allows the system to automatically adapt to changing service requirements while maintaining optimal resource utilization without increasing operational complexity.
3Reliability
If services are spread across multiple servers for redundancy, then service reliability improves, but resource allocation becomes fragmented and inefficient
Solution Approach 1:
The patent applies universality by enabling servers to serve multiple services and services to be hosted on multiple servers simultaneously through the auction mechanism. A single server can allocate resources to multiple winning bids from different service agents, and a service can bid on and receive resources from multiple server agents, creating efficient multi-server redundancy without fragmentation.
4Quantity of substance
If new servers are added to the resource pool to meet demand, then resource availability increases, but operating costs and power consumption increase
Solution Approach 1:
The patent applies parameter changes by dynamically adjusting the allocation parameters (bid amounts, resource quantities, prices) rather than physically adding or removing servers. The auction mechanism allows existing server resources to be reallocated to meet changing demands through parameter adjustments in the bidding process, avoiding the energy costs associated with provisioning and powering new hardware.
Data Source
AI summary
A method and apparatus for allocating server resources to services are provided. Multiple services compete for resources on one or more servers and bid for resources. Servers assign resources based on bids according to an auctioning rule mechanism. Services update bids according to a probabilistic dynamic that can approximate a continuous-time Markov chain, both for each service and for the collection of services. The method and apparatus can involve multiple separate but interacting agents, and can be implemented for example for self-organization of a datacentre. The behaviours of the agents can be configured so that the collective behaviour results in a proportional fair resource allocation.