Estimating Virtual Machine Capacity Using Allocation Flux
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Virtual Machine (VM) service providers face challenges in optimizing physical resource usage to mitigate costs associated with idle resources while ensuring adequate resources for current and future demands, as they need to forecast demand and make decisions with imperfect information about current resource usage levels.
Innovation Solution
A system that generates estimated remaining capacities (ERCs) for computing clusters using physical resource allocation flux data and VM type exchange probabilities, allowing for more cost-effective monitoring and decision-making without continuous, resource-intensive tracking of actual remaining capacities (ARCs) at the individual device level.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If actual remaining capacities (ARCs) are continuously monitored at the individual computing device level, then resource allocation accuracy is improved, but computing resource consumption and operational costs increase
Solution Approach 1:
The patent segments the monitoring task from the individual computing device level to the computing cluster level. Instead of tracking each device's ARC continuously, the system monitors aggregate cluster-level metrics and uses statistical methods to estimate individual device capacities. This segmentation reduces the overall monitoring burden and computing resource consumption while maintaining sufficient allocation accuracy.
Solution Approach 2:
The patent creates a virtual model (estimated remaining capacity) that copies the essential information needed for resource allocation decisions without requiring direct continuous measurement of actual remaining capacities. The ERC model replicates the functional need for capacity information while using less expensive monitoring approaches at the cluster level rather than individual device level.
2Reliability
If actual remaining capacities (ARCs) are continuously monitored, then VM deployment decision accuracy is improved, but system complexity and operational overhead increase
Solution Approach 1:
The patent introduces an intermediary layer (the ERC estimation system) between the physical monitoring infrastructure and the VM deployment decision-making process. This intermediary translates complex individual device monitoring requirements into simpler aggregate cluster-level metrics, reducing system complexity while maintaining the reliability needed for accurate deployment decisions.
Solution Approach 2:
The patent changes the monitoring parameters from detailed individual device ARCs to aggregate cluster-level metrics. By transforming the measurement parameters from fine-grained to coarse-grained, the system reduces operational overhead and complexity while still providing sufficient information for reliable VM deployment decisions through statistical estimation.
3Productivity
If physical resources are allocated to individual VMs with frequent monitoring, then resource utilization efficiency is improved, but the cost of tracking allocation flux increases
Solution Approach 1:
The patent implements periodic monitoring at the cluster level rather than continuous monitoring at the individual device level. By using periodic aggregate measurements and statistical methods to estimate individual device states, the system maintains effective resource utilization while significantly reducing the energy cost and operational burden of tracking allocation flux.
Data Source
AI summary
Predicting capacity of shared virtual machine (VM) resources by generating estimated remaining capacities (ERCs) for computing clusters within a virtualization system rather than continuously monitoring actual remaining capacities (ARCs). Generating ERCs for a virtualization system's computing cluster(s) by using physical resource allocation flux data and/or VM type exchange probabilities provides benefits over continuously monitoring ARCs. The physical resource allocation flux data may correspond to commissioning and decommissioning VMs into the cluster and may be obtained during a blind period when current ARCs are unknown. For example, the physical resource allocation flux may be an indication of how many instances of each VM type are commissioned and/or decommissioned from the cluster over a time interval of interest. The VM type exchange probabilities may indicate a likelihood that commissioning or decommissioning a VM type into the computing cluster(s) will take up or free up computing resources available to other VM types, respectively.