Decentralized Server Load Assignment for Data Center Energy Savings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for assigning computational loads to servers in data centers are inefficient, leading to high energy consumption and suboptimal server utilization, particularly in large and dynamic environments like Cloud Computing, where centralized solutions require constant updates and result in higher server maintenance and performance degradation.
Innovation Solution
A decentralized method using Bernoulli trials based on server load to determine server availability for virtual machine assignment and migration, allowing servers to autonomously decide on accepting or rejecting new loads, thereby reducing the need for centralized management and enabling scalable energy-saving strategies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If centralized solutions are used for VM assignment and migration, then optimal server utilization can be achieved, but system complexity increases and performance degradation occurs in large and dynamic data centers
Solution Approach 1:
The patent divides the centralized management function into distributed autonomous decisions at each server level. Each server independently performs Bernoulli trials to determine VM acceptance, eliminating the need for a complex centralized controller while maintaining optimization capabilities.
Solution Approach 2:
Servers autonomously make decisions about VM acceptance and migration using local Bernoulli trial mechanisms without requiring centralized coordination. This self-service approach reduces system complexity while achieving near-optimal server utilization through decentralized intelligent behavior.
2Loss of energy
If concurrent reallocation of multiple VMs is performed to optimize energy consumption, then energy saving increases, but service level and performance deteriorate
Solution Approach 1:
The patent implements sequential rather than concurrent VM reallocation by having each server independently perform Bernoulli trials and make decisions in a distributed manner. This periodic, staggered approach avoids the performance degradation and service level deterioration associated with simultaneous concurrent reallocation while still achieving significant energy savings.
3Loss of energy
If servers are consolidated to minimize active servers, then energy consumption decreases, but the number of active servers remains 20% higher than the theoretical minimum
Solution Approach 1:
The patent changes the decision-making parameter from deterministic threshold-based rules to probabilistic Bernoulli trial outcomes. This allows servers to make more nuanced decisions about VM acceptance and migration, achieving better consolidation efficiency with fewer active servers while maintaining service levels, thereby reducing energy consumption more effectively than previous methods.
Data Source
Figure 1~2
Figure 3
AI summary
The present invention concerns a method of assignment of computational load to the servers of a data centre by means of virtual machines aiming at the reduction of the energy consumed in said data centre and maintaining at the same time an appropriate service level for the clients of the data centre itself.