Resource Estimation

1. Reservations Statistical model predicts estimate Tolerance depends on medium latest start earliest start deadline tolerance estimate = mean + confidence Resource Estimation What Can You Do With It? http://research.microsoft.com/invisible Real-Time SOAP Self-tuning Planned Activities For Interoperating Low-Cost Devices Microsoft Research Johannes Helander, Stefan Sigurdsson Behavior Patterns What Is This? Define the actions and resources for a job Given to planner A programming model and software infrastructure for self-tuning coordination of embedded devices Technology applied Low cost microcontrollers (≈ $5) integrated into everyday objects using XML for data interchange Real-time scheduler on nodes Task planner delegates work Secure service discovery and adaptive resource allocation <behavior name="SensorDemo"> <action name="DemoInstigator“ endpoint="node:instigator/COB/sensormain.cob"> <message destination="SensorProducer/*"/> </action> <action name="SensorProducer“ endpoint="node:sensor/COB/sensor.cob"/> <repeat count="100" Period="P1.5S"/> <message destination="SensorConsumer"/> </action> <action name="SensorConsumer“ endpoint="node:consumer/COB/sensor.cob"/> <repeat count="100" period=”P1.5S”/> </action> <sampling destination="node:instigator“ interval="20" number="2"/> </behavior> Domain of interest Distributed real-time applications in ubiquitous computing scenarios Assisted living for the elderly Home and office automation Independent, zero effort deployment and interoperation across all device classes Pattern Instance: Task Specific time and place Sent to worker by planner App specifies confidence needed Higher confidence costs more <task name="SensorDemo-123456"> <action name="SensorProducer“ deadline=”2004-12-11T02:51:48.7001508Z” tolerance=”P0.005S” duration=”P0.02S”> <trigger maxCount="100“ offset="P1.5S">SensorProducer</trigger> </action> <sampling destination=http://10.10.10.10/feedback interval="20" number="2"/> </task> Planner Adaptation Feedback is collected by sampler Reservations adapt probabilistically Scheduling Fixed producer deadline Fixed consumer deadline Steady state Scheduling 1 1 Producer Time t Time t k·t0 k·t1 k·t2 k·t0 k·t1 k·t2 Sensor to recorder CD player to speaker Sampling Sensor readings Demo Structure Consumer Sensor scenario Measures real-world activity Heart rate, sound, temperature Produces readings at a regular rate Recorder action deadlines and durations are adjusted according to transmission time and measured processing requirements user mode I/O protocol driver sampler feedback PC PC probabilistic forecast planner serial-to- ethernet proxy serial line reservations real-time scheduler eb63 Speaker scenario As for the sensor scenario, except consuming must receive data at a regular rate, so producer adjusts sensor firmware SOAP/XML feedback sntp client real-time scheduler data readings data sink firmware SOAP/XML sntp client eb63 time server serial line PC Guarantee adequate resource availability and timely processing of critical tasks wired ethernet user mode I/O protocol driver serial-to- ethernet proxy