1 / 7

Phase Analysis and Prediction for Dynamic Resource Provisioning

Phase Analysis and Prediction for Dynamic Resource Provisioning. Jian Zhang and Renato Figueiredo ACIS Lab, University of Florida Mazin Yousif and Robert Carpenter Intel Corporation June 12, 2007. Motivation. Provide computing resources as a utility and charge the users for a specific usage

Download Presentation

Phase Analysis and Prediction for Dynamic Resource Provisioning

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Phase Analysis and Prediction for Dynamic Resource Provisioning Jian Zhang and Renato Figueiredo ACIS Lab, University of Florida Mazin Yousif and Robert Carpenter Intel Corporation June 12, 2007

  2. Motivation • Provide computing resources as a utility and charge the users for a specific usage • Exp. Amazon’s Elastic Compute Cloud (EC2) • User incentive: Request no more than the amount of resources that an application needs • Present a need to adapt the resource provisioning to the changing workload  minimize the costs to the users

  3. Problem Statement How to determine the time and space granularities of the resource allocation?

  4. Approach • Use virtual machines (VMs) as resource container to multiplex data center resources • Develop analytical tools to automatically discover the similarities and changes in the application’s resource consumption • Phase: A set of intervals within an application’s execution that have similar system-level resource consumption behavior • Phase analysis: Use k-mean clustering based algorithms to partition a set of intervals into phases and determine the best number of phases • Phase prediction: Use time series prediction and classification algorithms to predict future execution phases

  5. Application Containers Response Time VM Guest VM Guest … Performance Data VM Performance DB VMM Phase Analyzer Monitor Agent Training VM Host Phase Profile Application Phase DB Auditor CQ VM Configurations P1, P2,…,Pt ARM1 Prediction Accuracy Pt+1 Phase Predictor Resource Scheduler … ARMn VMM: Virtual Machine Monitor VM: Virtual Machine DB: Database ARM: Application Resource Manager CQ: Clustering Quality Pt: Phase ID at time t Learning Aided Application Phase Characterization Prototype

  6. Experimental Results (a) CPU_user of SPECseis96 (b) Bytes_out of WC’98 log replay Figure Phase Profile (Cp=8) • Cost Model: R(k) -- Resource cost TR(k) -- Re-provisioning cost P(k) -- Mis-prediction penalty P(k) C and Cp -- The relative unit cost of phase transition, and mis-prediction penalty with regard to the unit cost of resource reservation

  7. Thank you

More Related