Queuing Model for Web Workload Characterization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing resource allocation methods in online systems, such as data centers, struggle to accurately predict web workload fluctuations and account for time-varying traffic intensities, leading to impractical trial-and-error approaches in system management and capacity planning.
Innovation Solution
A method is developed to generate a queuing model using a Markovian Arrival Process (MAP) for web servers, which parameterizes arrival and service times to predict server performance, handling batch requests and different request types, and computes performance metrics like queue length and response time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional trial-and-error methods are used for resource allocation, then system management tasks can be performed, but the process becomes impractical as data centers grow larger and workloads become more complex
Solution Approach 1:
The patent creates simplified computational models (queuing models) that copy and represent the complex behavior of large data center systems. These models capture essential workload characteristics and server performance dynamics without requiring full-scale simulations, enabling practical resource allocation decisions for growing systems.
Solution Approach 2:
The patent transforms complex workload characteristics into manageable model parameters through automated fitting procedures. By changing the representation from raw trace data to fitted model parameters, the system can efficiently handle increasing complexity without requiring proportional increases in computational resources or time.
2Measurement precision
If detailed trace data analysis is performed to accurately characterize web workload, then prediction accuracy improves, but the computational complexity and time requirements increase
Solution Approach 1:
The patent performs preliminary fitting of queuing models to trace data during an offline phase, creating pre-computed models that capture workload characteristics. These pre-fitted models can then be used for rapid performance prediction without requiring real-time analysis of detailed trace data, significantly reducing online computational time.
Solution Approach 2:
The patent segments the workload characterization process into distinct phases: trace data collection, model fitting, and performance prediction. By separating the computationally intensive fitting phase from the prediction phase, the system achieves both accurate characterization and efficient runtime performance.
3Device complexity
If conventional queuing models are used that assume Poisson arrivals and exponential service times, then the models are simple to implement, but they fail to accurately represent real web workload patterns including batch requests and different request types
Solution Approach 1:
The patent generalizes conventional queuing models by changing the arrival process from Poisson to Markovian Arrival Process (MAP) and service time distribution from exponential to general distributions. This parameter change allows the models to capture complex workload patterns like batch requests and different request types while maintaining the analytical tractability needed for practical implementation.
Solution Approach 2:
The patent combines multiple modeling components (MAP for arrivals, general service time distributions, state-dependent service rates) into a composite queuing model. This composite approach integrates various features to accurately represent real web workloads while preserving the mathematical structure needed for performance analysis.
Data Source
AI summary
Implementations of the present disclosure provide systems and methods directed to receiving, at a computing device, trace data provided in a trace log file corresponding to a server, parameterizing, using the computing device, a first Markovian Arrival Process (MAP) model based on the trace data, parameterizing, using the computing device, a second MAP model based on the trace data, defining a queuing model that models a performance of the server and that includes the first MAP model and the second MAP model, and storing the queuing model in computer-readable memory.