Motivation for Studying Adaptive Middleware Flexible Scheduling in Middleware for Trends Distributed Rate-Based Real-Time •Hardware keeps getting smaller, faster, & cheaper Applications •Software keeps getting larger, slower, & more expensive Christopher D. Gill Historical Challenges AbstractService service •Building distributed systems is hard Client Dissertation Supervisors: Dr. Ron K. Cytron, Dr. Douglas C. Schmidt •Building them on-time & under budget is even harder Proxy Service 1 1 Department of Computer Science service service Washington University, St. Louis, MO New Challenges cdgill@cs.wustl.edu •Many mission-critical distributed applications require real-time QoS guarantees – e.g., combat systems, online trading, telecom •Building QoS-enabled applications manually is tedious, error-prone, & expensive •Conventional middleware does not support real- time QoS requirements effectively Tuesday, December 18, 2001 Overview of Research Areas Motivating Applications Technical Research Research Boeing Bold Stroke Middleware Challenge Approach Impact Infrastructure Platform Efficient and Safe Systems •Operations well defined •Event-mediated middleware solution Customizable Middleware •Previous-generation systems static Nimble Adaptation •Next-generation systems dynamic and adaptive Performance Awareness Domain Company Avionics mission computing Boeing, Raytheon Inclusive Systems Mass storage devices SUTMYN, StorTek Approach Medical Information Systems Siemens, GE Satellite Control LMCO COMSAT Telecommunications Motorola, Lucent, Nortel, Cisco, Siemens Missile & Radar Systems LMCO Sanders, Raytheon These technical challenges raise crucial systems issues These technical challenges raise crucial systems issues Steel Manufacturing Siemens ATD in both theoretical and empirical dimensions in both theoretical and empirical dimensions Beverage Bottling Automation Krones AG Limitations With Existing Approaches Research Contributions Systems Architecture and Framework Historically, each application has solved scheduling on its own • Extends open-source middleware • Tedious, error-prone • Costly over system lifecycles Patterns for Adaptive Scheduling • Single-paradigm approaches • Capture design experience and solutions Current middleware lacks hooks for key Empirical Evaluation of Adaptive Scheduling domain-specific features, e.g. : • Practical benefits for real-world applications • Optimized integration with higher level managers • Hybrid static-dynamic scheduling strategies • Strategies built from primitive elements • Adaptive domain-specific optimizations 1
Research Approach: the Kokyu Flexible Middleware Scheduling/Dispatching Framework Integration with Event Service Scheduler rate propagation Dispatcher Dispatching WCET propagation configuration • Separates dispatching selected propagated mechanism from rates rates static rate RMS scheduling policy tuples static • Dispatcher consults run- mandatory time scheduler for priorities optional LLF tuple laxity • Flexibility for different visitor sub-graph timers scheduling policies operation Rate and priority visitors assignment policy Dispatcher is (re)configurable Dispatcher is (re)configurable • Multiple priority lanes • Multiple priority lanes • Queue, thread, timers per lane • Queue, thread, timers per lane Application specifies characteristics Application specifies characteristics • Starts repetitive timers once • Starts repetitive timers once • e.g., criticality, periods, dependencies • e.g., criticality, periods, dependencies • Looks up lane on each arrival • Looks up lane on each arrival Scheduler assigns rates & priorities Scheduler assigns rates & priorities Implicit projection Implicit projection per topology, scheduling policy • Of specific scheduling policy into per topology, scheduling policy • Of specific scheduling policy into generic dispatch infrastructure generic dispatch infrastructure • Defines necessary dispatch configuration • Defines necessary dispatch configuration Safety: Meet Critical Deadlines Greater Utilization with Criticality Isolation Dynamic Static Technical Research Research Challenge Approach Impact Efficient and Safe Arbitrary strategies that Increased utilization, Systems hybridize static/dynamic critical operations still scheduling/dispatching meet their deadlines Customizable Middleware Nimble Adaptation Performance Awareness Inclusive Systems Approach same point in execution Solution: Support for Hybridizing Static & No One Strategy is Optimal Dynamic Scheduling Heuristics Technical Research Research Challenge Approach Impact Dispatcher Dispatching Efficient and Safe Arbitrary strategies that Increased utilization, configuration Systems hybridize static/dynamic critical operations still Abstract mapping: operation scheduling/dispatching meet their deadlines characteristics → OS & Customizable Middleware Dispatching composed Supports tailored “fit” of middleware primitives laxity from primitive elements scheduling/dispatching • Generalizes RMS, EDF, LLF, MUF, MUF RMS+LLF, RMS+EDF, … Nimble Adaptation laxity • Arbitrary composition of primitives • Drives factory-driven dispatching module (re)configuration at run-time Performance Awareness (work in progress) critical non-critical Inclusive Systems Approach 2
Solution: Compose Scheduling Heuristics from Case for Multi-Paradigm Scheduling Dispatching Primitives Dispatcher ) static feasibility static laxity RMS MUF static laxity +LLF performance static laxity Dynamic1 mandatory optional Dynamic2 Static Gives Fine Grain Control over Feasibility / Performance Trade-Off Solution: Kokyu Real-Time Metrics Unobtrusive Monitoring and Control Feedback Monitoring Framework Technical Research Research Challenge Approach Impact SHARED MEMORY Consistent View of Time Efficient and Safe Arbitrary strategies that Increased utilization, REMOTE LOGGER Systems hybridize static/dynamic critical operations still METRICS CACHE scheduling/dispatching meet their deadlines Customizable Middleware Dispatching composed Supports tailored “fit” of S Customized Data Collection E STORAGE B from primitive elements scheduling/dispatching O R METRICS P REMOTE WORKSTATION O(n 2 )/O(n log n) → MONITOR Nimble Adaptation Integrated rate/priority PROBES OPERATIONS selection mechanisms EMBEDDED BOARDS O(n log n)/O(n) bound Logging and Visualization on adaptation DISPATCHER Performance Awareness Time and space Provides run-time DOVE efficient data collection observable info for Browser QuO (Java) Feedback to Higher-Level and storage framework control, post-analysis Resource Managers Inclusive Systems RTARM FRAME Approach MANAGER Experimental Test-Bed Popular Scheduling Strategies Application • Research Version of Operational Flight Program for AV-8B Aircraft • Added navigation route computations to ramp non-critical load Rate Monotonic Scheduling (RMS) • Added critical and non-critical computations to inject execution time jitter • Assigns thread priorities by rate • Operations at each priority handled in FIFO order Middleware • The ACE ORB (TAO) • TAO Real-Time Event Channel RMS + Minimum Laxity First (MLF) • Kokyu Framework: Scheduling, Dispatching, and Metrics • Critical operations managed as in RMS • Non-critical operations managed in single lowest priority Operating System and Hardware • Non-critical operations handled in minimum laxity (slack time) order • VxWorks RTOS on the PPC boards • 200 MHz Motorola PPC 604 card Maximum Urgency First (MUF) • two 100 MHz Dy4-177 PPC 603 cards • Thread priority per criticality level • Dy4-783 memory mapped display processor • Operations in each priority level handled in laxity order • Commercial VME-64 chassis with all boards • Switched Ethernet, MIL-STD-1553 MUX Bus on one Dy4-177 card 3
Recommend
More recommend
