jade
play

JADE An adaptable infrastructure to build autonomic system Sara - PowerPoint PPT Presentation

JADE An adaptable infrastructure to build autonomic system Sara Bouchenak, Noel De Palma, Daniel Hagimont, Sacha Krakowiak Jean-Bernard Stefani SARDES Project, INRIA RHONE ALPES, GRENOBLE, FRANCE 1 Outline Motivations Design


  1. JADE « An adaptable infrastructure to build autonomic system » Sara Bouchenak, Noel De Palma, Daniel Hagimont, Sacha Krakowiak Jean-Bernard Stefani SARDES Project, INRIA RHONE ALPES, GRENOBLE, FRANCE 1

  2. Outline � Motivations � Design principle � Use case � Jade for clustered J2EE application � Deployment and Repair management � Performance � Conclusion 2

  3. Motivation � Distributed Software � Complex, heterogenous and legacy � Management is required but nightmarish � Example of management function � Software installation � Software configuration � Performance Tuning � Fault Tolerance � Securité 3

  4. problems statement � Management software � It can be a complex distributed application � Management � Complex task � Achieved by human � Consequence � Error (mainly configuration) � Low reactivity � Consume a lot of resources � human resources � hardware resources (overbooking) 4

  5. Approach : autonomic management software � Management software � Installation, deploy, configure … � The managed system (Legacy) � The management system itself � Autonomic behavior � monitor, decide, reconfigure � The Legacy application � The management system itself � Benefit � Less error � More reactivity � Resources friendly � We need the same abstraction � The managed system � The management system 5

  6. Jade : design principle � Component model (Fractal - Julia - Think) � To model and wrap legacy managed resource � Component abstraction Component abstraction over over legacy legacy software software � � To build the management software � Jade is adaptable and jade is self managing � Autonomic Autonomic manager manager works works on component abstraction on component abstraction � Jade Jade � Management software Autonomic Managers Autonomic Managers system representation system representation � Bootstrap (self-deployable) Failure Manager Failure Manager Jade Jade � Autonomic Manager Autonomic Autonomic Managers Managers � Managed Resource SR SR Managed System Managed System � Explicit Control Loop Managed System Managed System Tomcat � System representation Apache MySql Tomcat 6

  7. Viagra is Fault-tolerant � A Jade personality to manage clustered J2EE application � Repair management � The application � RUBiS (servlet version based on e-bay model) Presentation tier Database tier Web tier mod_jk plugin Tomcat Users AJP13 JDBC L4-switch - Tomcat 7

  8. Jade in this context � Component abstraction is used to model/to control � Virtual cluster � Middleware � Application Naming Global Software Service Manager repository Cluster Failure Application Manager Manager Manager System representation 8

  9. Focus on the failure manager Repair management � Component based Component based system’s architecture (Fractal) � � Repair of a components structure (legacy application or jade itself) � Fail stop Fail stop failure of node � � Configurable Configurable and generic generic repair policy � � Updating the failed managed system conform to the configuration in place prior to the occurrence of failure 9

  10. Failure Manager � Explicit control loop Explicit control loop Failure Manager � Sensor Repair � Actuator � Repair policy sensors actuators � Manifests the relationship between managed system and repair management functions � Repair policy Repair policy described through Managed System Managed System an ADL ADL � Set of actuators � Actuators scheduling Self configurable � Self configurable � 10

  11. Evaluation Environment: Deployment Architecture RUBiS Apache MySQL Client Emulator Web Server Database 11 Tomcat Servlet Containers

  12. Evaluation Setup � RUBiS transition table � requests � 600 users (TPCW think time) � Test time � 500s ramp-up time � 4000s test time � 500s down-ramp time � 3 consecutive Tomcat crashes (distance 1000s) 12

  13. Expected / Ideal Graphs response time Unmanaged System time response time Managed by JADE time 13

  14. CPU Usage – no JADE 14

  15. CPU Usage - JADE 15

  16. conclusion and future work � Management achieved by human � Error (mainly configuration) � Low reactivity � Consume a lot of resources � Our approach : Autonomic Management � Use Case and first evaluation J2EE � Failure manager � � Under development � Autonomous Self sizing for J2EE applications � Next � Autonomic management of message base application � Fractal, Julia, Think are in LGPL � JADE is soon in LGLP 16

Recommend


More recommend