05/06/2003Lowell No Knobs Dave DeWitt and Dieter Gawlick
05/06/2003Lowell A List of Issues Auto Tuning Trade-off decisions Selection of functionality Discovery
05/06/2003Lowell Auto Tuning Dynamic allocation of existing resources CPUs (blades), CPU cycles, processes, memory, storage, bandwidth, … Index management, materialized view, query optimization, query re-write, … Integrated with failure reaction (Disposable disks (RAID), disposable computers (blades)) Low variance in response time Advisors for need/impact of new resources
05/06/2003Lowell Auto Tuning Single/multiple database(s)/database instance(s) Two computer centers Few blade computers in each This is mostly engineering work Keep knobs Make use optional, maybe set by/through applications Prepare code Monitor, adjust All database vendors are working hard on it
05/06/2003Lowell Auto Tuning Research questions? What to do with a bad query/program? Reports are not the solution 80% case: How do we know that we are at 80%? Maybe we are at 1/10% of optimum
05/06/2003Lowell Trade-off Decisions Which application/task/user is most important? To be defined on high level Rules and regulation Generic applications should set this should set this Business value of application Business transactions or part thereof Generic Content dependent Organization People (static resources)
05/06/2003Lowell Trade-off Decision To be used as input for auto tuning A model for high level abstractions would be helpful - Research issue?
05/06/2003Lowell Selection of Functionality Feature Enabling Option Customization, preferences Sources Rules and regulations Domain, enterprise, department, people (Security) Should be closely tied to model for trade-off decisions
05/06/2003Lowell Discovery Low level discovery process New/broken disks, blades, … Depending on naming conventions and standards Mostly a systematic engineering effort High level discovery process Semantic transformation, …