70 likes | 198 Views
Component Level Performance Monitoring Of Enterprise Applications in Cloud Data-centers. Shankaranarayanan P N, Ashiwan Sivakumar. Ou r Research Focus. Enterprises moving applications to the cloud. Cost savings, geo-redundancy.
E N D
Component Level Performance Monitoring Of Enterprise Applications in Cloud Data-centers Shankaranarayanan P N, Ashiwan Sivakumar
Our Research Focus • Enterprises moving applications to the cloud. • Cost savings, geo-redundancy. • Enabling performance aware enterprise applications on the cloud. • e.g., Enterprise Resource Planning, Supply Chain Management. • Latency and availability are critical. • Complex, multi-tiered, large-scale. • Understanding sources of variability with performance in the cloud. 2
Why is the problem difficult? • Amazon’s EC2 Cloud Titanic went down on April 22 2011 • Northern Virginia data center – domino effect with EBS • Microsoft’s online services went down on Sep 9 2011. • DNS address not propagated. • Google docs down on Sep 7 2011 – same reason • Variations in infrastructure and failures – unpredictable. • Availability and Closer to Users – have multiple instances 3
L6,4 CS(10) CS(9) Deployment in the cloud – Components view L5,2 L1,9 L9,4 L6,10 L2,9 FE(1) L7,4 L5,10 L2,4 BS(6) BS(2) DB(4) L1,2 User(0) Data center 1 L10,4 L2,3 L3,4 L6,7 FE(5) Data center 2 OS(7) OS(3) L5,6 4
Challenges • Obtaining and collecting data from the performance measurements made in cloud data centers. • Analyzing the data to get the required information • Building a generic and application agnostic measurement infrastructure • Estimating the component capacity in the cloud dynamically. 5
Approach to solution • Profiling Framework • Leverage infrastructure level performance counters • Mapping infrastructure data with user performance • Use tracing tools like Xtrace • Apply techniques like Aspect Oriented Programming • Dynamic capacity estimation • Probing at regular intervals similar to TCP slow start 6
Q&A 7