230 likes | 326 Views
An Active Events Model for Systems Monitoring. Philip Gross Columbia University Programming Systems Lab Director: Gail Kaiser. Motivation: Loose Coupling. Want to maximize ease of distributed programming Not only should middleware deliver events…
E N D
An Active Events Model for Systems Monitoring Philip Gross Columbia University Programming Systems Lab Director: Gail Kaiser
Motivation: Loose Coupling • Want to maximize ease of distributed programming • Not only should middleware deliver events… • It should completely shield participants from knowledge of each other • Information goes into the cloud, information comes out of the cloud 2
Problem: No Event Context • Consumers get data that they’re “supposed” to get • Confusion and ambiguity can be major problems 3
Example: Stock Trading Program • Sub (MSFT > 80) // sell and cash in • Event published with MSFT = 60: ignored • Event published with MSFT = 1515: received, all shares sold • Are we rich yet? • Unfortunately no: was appointment with MS sales rep at 3:15pm 4
We Conclude That… • We need to associate semantics with event data • Somehow saying “this is what I mean,” or “here’s how to interpret me” • Events that explicitly indicate semantics, we call Active Events, or ActEvents • They’re not just passive data containers 5
Two Flavors of Information Bus • Light model: information is carried by a notification service • High speed, low latency, simple behavior • Heavy model: information is carried by self-transmitting software agents • Lower speed, higher latency, more sophisticated behavior 6
And Two Flavors of ActEvents • By reference or by value • By reference: SmartEvents • Data includes structured tags • External processor maps tags to semantics • By value: Gaugents • Agent ferries not just data… • But also mobile code that has semantic knowledge 7
SmartEvent Implementation • XML embedded in event • Certain tags “exposed” for routing • XML annotated to provide syntactic and semantic information • FleXML (Flexible XML) enables dynamic runtime schema and semantics discovery and composition 8
Initial SmartEvents • Language recognition problem • Validate against Schema • When unknown element encountered, ask “Semantic Server” for • New schema fragment • Processing modules (Tag Processors) • Tag Processors give higher-level events • Supports dynamic schema composition 9
Newer SmartEvents • XML is described with multiple namespaces • Namespaces provide direct syntactic identification • Which allows lookup of Tag Processors • Semantics still looked up dynamically • Still supports dynamic schema composition • Tag Processors can maintain state over event streams 10
Two Flavors of Information Bus • Light model: information is carried by a notification service • High speed, low latency, simple behavior • Heavy model: information is carried by self-transmitting software agents • Lower speed, higher latency, more sophisticated behavior 11
Gaugent Implementation • Specialization of Worklets mobile agents • Worklet Virtual Machine (WVM) installed at each participating node • Java bytecode based • Routing can be altered on the fly • Carries data and processing code 12
Gaugents Continued • Worklets normally carry one junction for each WVM they plan to visit • Encapsulates computation to be performed at that location • Contains semantic information • May carry jackets • Provide a mini-workflow for a particular junction, e.g. repetition, start/exit constraints, etc. • Contain manipulable routing information 13
Future Directions • New event system • Extended FleXML • Web services 16
Future Event System • MEEP? CHIEF? • Native XML • Scalability, survivability, performance • VPENs: Virtual Private Event Networks • Event diodes for Multi-Level Security • Content-based routing, but with significant optimizations for topic-based advertisements • GPL 17
Conclusions • Semantic annotation of events is essential for loosely coupled architectures • Different solutions are appropriate for different types of information exchange • We have built two implementations • More experimentation needed 18
Questions? 19
Extended FleXML • Currently one interpretation per incoming event per Metaparser • Multiple consumers may need multiple interpretations • Thus sets of Tag Processors may be applied to produce multiple high-level “interpretations” 20
Web Services • Juggernaut: Sun, IBM, Microsoft all pushing it hard • Has significant implications for service interoperability projects • In particular: SOAP and WSDL 21
SOAP • Very nice encapsulation • Future probe-gauge standard? <env:Envelope xmlns:env="http://www.w3.org/2001/09/soap-envelope"> <env:Header> <n:alertcontrol xmlns:n="http://example.org/alertcontrol"> <n:priority>1</n:priority> <n:expires>2001-06-22T14:00:00-05:00</n:expires> </n:alertcontrol> </env:Header> <env:Body> <m:alert xmlns:m="http://example.org/alert"> <m:msg>Pick up Mary at school at 2pm</m:msg> </m:alert> </env:Body> </env:Envelope> 22
WSDL • Web Services Description Language • Useful for where Pub/Sub meets RPC • Types/Messages • Operation: organize into inputs and outputs • PortType: collection of Operations • Binding: mapping of PortType to e.g. SOAP • Service: set of bound PortTypes at an actual location 23