Project Proposal “Profile-driven Qo. S-aware Online Services” by Amitayu Das Sri Hari Krishna Narayanan CSE 598 B Fall 2005
Problem in Qo. S satisfaction • Demand resource requirement allocated resource • Application characteristics • Time-varying demand, transient overload • Limited resource • Dynamic system management policies/mechanisms
Resource-profiling • How resource is being used by application(s): act of capturing that data • What sort of data? (workload, appln, system, architecture parameters etc. ) • Collect data at regular intervals • Mapping: f (λwl, δwl) == ==> (θcpu, θmem, θdisk)
How does it help? • Profiling helps in understanding application characteristics • Helps in better resource allocation, better control • Helps in prediction of performance • How to decide optimal interval-length? • How to decide optimal interval-length online?
Objective • Evolve to decide about the optimal intervallength in a self-tuning manner • Different time-points may have different optimal interval-length: figuring that out • Allocate resource based on new profiles to satisfy Qo. S: self-tuning performance management
Strategies • Offline determination: – – Try to capture “significant change” Get profiles with different interval-length Define metric to compute utility of a profile Decide the best interval-length from best profile • Online determination: – Figure out if overload of offline method is tolerable or not – If not, then come up with other mechanism • Resource Allocation: not yet decided
Plan (time-frame) • Literature survey: throughout • Offline measurement: – Set up system (kernel instrumentation): 2 weeks – Data collection: 3 weeks – Analysis: 2 weeks • Online measurement: – Check for the overload: 2 weeks – Formulating other strategies: undetermined • Resource allocation: depends on above
References • “Performance Modeling and System Management for Multi-component Online Services” by C. Stewart et al. • “An Automated Profiling subsystem for Qos-aware Services” by T. Abdelzaher • “Load Profiling in Distributed Real-time Systems” by A. Bestavros • “Measuring and Characterizing System Behavior Using Kernel-Level Event Logging” by K. Yaghmour et al. • “Resource Overbooking and Application Profiling in Shared Hosting Platforms” by B. Urgaonkar et al.
• Questions ? ? ? • Thank You !!!
Скачать презентацию Project Proposal Profile-driven Qo S-aware Online Services by
2110127ba9220ca2a5970309f5be6a11.ppt