1 / 21

Dynamic Resource Allocation for Shared Data Centers Using Online Measurements

Dynamic Resource Allocation for Shared Data Centers Using Online Measurements. Abhishek Chandra Weibo Gong Prashant Shenoy UMASS Amherst http://lass.cs.umass.edu/projects/shop. Motivation. Data Centers Server farms Rent computing and storage resources to applications

mbanda
Download Presentation

Dynamic Resource Allocation for Shared Data Centers Using Online Measurements

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. Dynamic Resource Allocation for Shared Data Centers Using Online Measurements Abhishek Chandra Weibo Gong Prashant Shenoy UMASS Amherst http://lass.cs.umass.edu/projects/shop

  2. Motivation • Data Centers • Server farms • Rent computing and storage resources to applications • Revenue for meeting QoS guarantees • Goals: • Satisfy application QoS guarantees • Maximize resource utilization of platform • Robustness against “Slashdot” effects

  3. Dynamic Resource Allocation • Periodically re-allocate resources among applications • Estimate resource requirements for near future • Challenges: • Reallocation at short time-scales • No prior workload profiling/knowledge • Low overhead • Approach: Online Measurement-based Allocation

  4. Talk Outline • Motivation • System Model • Dynamic Allocation Techniques • Experimental Results • Conclusions

  5. Resource Model • Queuing System • Generalized Processor Sharing (GPS) scheduler • Request classes • Different arrival processes, service time distributions • QoS Goal: Mean Response Time Resource GPS

  6. Expected Load APPLICATION MODELS PREDICTOR Measured Usage Rsrc Reqmts MONITOR ALLOCATOR System Metrics Resource Shares RESOURCE Dynamic Resource Allocation

  7. Expected Load PREDICTOR Measured Usage MONITOR System Metrics RESOURCE Dynamic Resource Allocation APPLICATION MODELS ALLOCATOR

  8. Adaptation Window History Measurement Interval Monitoring • Measure system and application metrics • Queue lengths • Request response times • Monitoring windows Time

  9. Mean AR(1) Last value History Adaptation Window Prediction • Short-term prediction of workload characteristics • Request arrival rate • Average service time • Use history of measured system metrics

  10. Workload Prediction Prediction Error Time (min) Prediction Accuracy

  11. Expected Load APPLICATION MODELS Rsrc Reqmts Resource Shares RESOURCE Dynamic Resource Allocation PREDICTOR MONITOR ALLOCATOR

  12. Measurement-based Model • Goal: Relate QoS metric to resource requirement • Idea: Model parameterized by online measurements • Advantages: • Parameters do not need to be computed • Allow adaptation to dynamic workload • Proposed: Transient Queuing System Description

  13. Transient Queuing Model • Transient queuing behavior over adaptation window • Relation between mean response time T¯ and application share w • Little’s Law: • Relation is parameterized by the measured workload • Arrival rateλand mean service time s¯

  14. Resource Allocation: Utility Model • Discontent function: Measures the QoS violations of an application • Constrained Optimization problem u1 Optimization u2

  15. Discontent Di Goal Response Time subject to Constrained Optimization Formulation • Non-linear Optimization Problem: • Solved using Lagrange multiplier method

  16. Talk Outline • Motivation • System Model • Dynamic Allocation Techniques • Experimental Results • Conclusions

  17. Experimental Setup • Simulation experiments • Soccer World Cup’98 Traces • Results based on a 24-hour portion of the trace • 755,000 requests • Mean req rate: 8.7 req/sec • Mean req size: 8.47 KB

  18. Workloads Share Allocation Adaptation to Transient Overloads Shares adapt to changing workload characteristics

  19. Adaptation: System Discontent GPS without reallocation GPS with reallocation System Discontent is lowered substantially

  20. Conclusions • Dynamic Resource Allocation needed for data centers • Measurement-based allocation: • Monitoring and Prediction gather online state • Use this state for application modeling and allocation • Future Work: • Prediction policies • Utility functions http://lass.cs.umass.edu/projects/shop

  21. Related Work • Prediction • Statistical Prediction Models [Zhang00] • Application Models • Queuing-Theoretic Models [Carlstrom02,Liu01] • Control-Theoretic Models [Abdelzaher02,Lu01] • Data Centers • MUSE [Chase01] • COD [Moore02] • Oceano [Appleby01]

More Related