Queuing Model for Virtualized Storage Contention Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In virtualized data centers, predicting the performance degradation due to storage device contention among multiple Virtual Machines (VMs) is challenging due to complex interactions of I/O flows, caching mechanisms, scheduling algorithms, and communication protocols, which affects latency and management decisions.
Innovation Solution
The method involves generating traces of I/O requests from isolated VM environments, parameterizing a queuing model to represent the consolidated virtualized environment, and using this model to estimate response times by simulating the queuing network with fair share scheduling, accounting for request splitting and merging operations at the Virtual Machine Monitor (VMM) level.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple VMs share a storage device in a consolidated virtualized environment, then resource utilization improves, but performance degradation occurs due to I/O contention
Solution Approach 1:
The patent creates a simplified predictive model that copies the essential characteristics of the complex virtualized I/O system. Instead of simulating the entire VMM, caching mechanisms, and scheduling algorithms, the model captures key parameters like I/O request patterns and storage device performance to predict contention effects without requiring full system complexity
Solution Approach 2:
The patent introduces a predictive model as an intermediary between the complex virtualized environment and performance analysis. This model acts as a mediator that translates complex I/O interactions into predictable performance metrics, enabling administrators to estimate contention effects without direct measurement in the consolidated environment
2Measurement precision
If detailed modeling of I/O flows and caching mechanisms is implemented, then prediction accuracy improves, but model complexity increases
Solution Approach 1:
The patent extracts only the essential elements needed for accurate prediction from the complex virtualized I/O system. Rather than modeling all components (VMM, caching, scheduling algorithms, device drivers), it isolates and models the critical factors that dominate performance: I/O request patterns from VMs and storage device response characteristics
Solution Approach 2:
The patent transforms the complex modeling problem into a parameter-driven approach. Instead of simulating system behavior through complex interactions, it uses key parameters (I/O workload characteristics, storage device performance metrics) to predict outcomes, simplifying the model while maintaining accuracy through focused parameter selection
Data Source
AI summary
Implementations of the present disclosure provide computer-implemented methods for predicting a performance of a consolidated virtualized computing environment. Methods include processing benchmark workloads using a plurality of virtual machines to generate a plurality of traces, each trace including data corresponding to requests issued by a respective virtual machine operating in an isolated environment, storing the plurality of traces in one or more trace repositories, each trace repository provided as a computer-readable storage medium, selecting a trace from the plurality of traces stored in the one or more trace repositories, parameterizing a queuing model based on the trace, the queuing model representing request queuing in the consolidated virtualized computing environment, and processing the queuing model using one or more processors to generate one or more response time estimates for the consolidated virtualized computing environment.