Adapting Applications and Platforms Vertical Customization for Event-based QoS Dr. Karsten Schwan, Dr. Greg Eisenhauer Ada Gavrilovska, David Robinson, Qi He Application Component Host Extension Modules Active NI Middleware Extensible Platforms OS Adapting Applications and Platforms Quality Attributes Project Goals IQ-ECho’s middleware enables integrated QoS management through quality attributes. Attributes implement the cross-layer information exchanges needed to make effective runtime tradeoffs in quality vs. performance across the protocol, middleware, and application levels. Configurable protocols and kernel-level monitoring provide the system-level support required for online quality management. Evaluation/integration with wide-area applications: remote visualization/ sensors, access/data grid; wireless systems IXP Approach Host Fast buffer-based host exchange protocol Fast-path Reliable UDP protocol on NIC On-the-fly code generation for moving application logic to NIC IXP Active NI RUDP Dyn. Code Network IQ-ECho Approach: Adaptive Applications on Programmable Platforms ECho: High performance events: Event-based peer-to-peer data communication infrastructure QoS: Runtime adaptation via dynamic code generation and mirroring Source-based filtering: meet required application QoS by disposition of application-specific code into remote sites and underlying platform Mirroring: utilize redundancy to improve responsiveness IQ-ECho: Integrated QoS management through dynamic quality attributes: Adaptation of platform (protocols and middleware) and of applications (adaptation-awareness) Publications and Demos Supercomputing 2000 efficient binary message formats Int. Parallel and Distrib. Processing Symposium (IPDPS) 2000 active User Interfaces High Performance Distributed Computing (HPDC) 2000 ECho Real-Time Applications and Systems (RTAS) 2001 Quality Events High Performance Distributed Computing (HPDC) 2001 Mirroring, Filtering, XML Supercomputing 2001 Path-based Filtering) Demo Differential Filtering QoS: Source-based Filtering: Achieving Desired QoS via Dynamic Code Generation Source F( ) Sink F( ) Source Filter Function: 2500 { if ((data inside viewing area) and (data values inside useful ranges)) { return 1; /* submit into channel */ } return 0; /* suppress data */ 2000 1500 1000 500 Active Sink Passive Sink Multiple Ch Single Ch Atl <-> Mun