Successfully reported this slideshow.

Selfxl Project Solutions Linux Ow2

1,028 views

Published on

  • Be the first to comment

  • Be the first to like this

Selfxl Project Solutions Linux Ow2

  1. 1. SelfXL project Autonomic management of PaaS cloud infrastructure with JASMINe Jean-Marc Menaud, Ecole des Mines de Nantes Benoit Pelletier, Bull Frédéric Dang Tran, Orange Labs This research is supported by the french Agence Nationale de la Recherche with the ANR-08-SEGI-017 18 Mars 2010
  2. 2. SelfXL  Self-management of compleX and Large scale systems  Acquire new processes in the field of the autonomic computing © OW2 Consortium 2010 http://selfxl.gforge.inria.fr 2
  3. 3. Partners  Bull  Open Source Java EE Middleware R&D  Virtualization system expertise  France Telecom / Orange Labs  R&D expertise in distributed systems architecture, management of virtual IT infrastructures  Mines de Nantes  Virtualization and constraint programming expertise.  IRIT  Meta-modelisation expertise.  LIG  Data mediation  ScalAgent  Mediation system © OW2 Consortium 2010 http://selfxl.gforge.inria.fr 3
  4. 4. Cloud Computing at France Telecom  Context:  Large-scale datacenters (+40,000 x86 servers)  JavaEE JOnAS middleware stack: +250 applications, +1000 application server instances  Static server consolidation (12/1) through virtualization: • Utilization level increased from 20% to 75%  Next step: migration to a Cloud model  On-demand JavaEE server provisioning according to demand fluctuations  Automation of the management of virtualized JavaEE clusters: deployment, update, auto-scaling etc.  Open (source) cloud management stacks alternatives to proprietary solutions (IaaS/PaaS) © OW2 Consortium 2010 http://selfxl.gforge.inria.fr 4
  5. 5. Self-sizing for Java EE PaaS  Goal : improve the platform efficiency (performance, energy cost)  Challenges  Cluster growth/shrink capabilities according to the load  Datacenter multi-tenancy with limited capacities requires arbitration policies  Datacenter placement for minimizing the energy consumption © OW2 Consortium 2010 http://selfxl.gforge.inria.fr 5
  6. 6. Architecture for self-sizing Scale up/down VM provisioning VM placement © OW2 Consortium 2010 http://selfxl.gforge.inria.fr 6
  7. 7. JASMINe  Management tools for SOA platform  OW2 Open Source project (license LGPL, Leadership Bull)  Cross-platform (JOnAS AS prime target)  Four main features (Design, Deploy, Monitoring, Self-*) © OW2 Consortium 2010 http://selfxl.gforge.inria.fr 7
  8. 8. JASMINe & MAPE loop  Monitor  JMX probes (JavaEE, Hypervisor)  Mediation bus  Graphs console  Analyse/Plan  Rules engine  Execute  Reconfiguration load-balancer API  JASMINe Virtual Machine Management (VMM) for VM provisioning & deployment © OW2 Consortium 2010 http://selfxl.gforge.inria.fr 8
  9. 9. CLIF is a load injection framework  CLIF is a software framework for load testing  load injectors (workload generation and response time measurement) and probes (resources usage monitoring)  open, adaptable/extensible, independent from: • System Under Test (protocols, probes/observed resources) • workload definition mode • user interfaces (Java/Swing, Eclipse, command line...)  distributed infrastructure designed for high level workloads  100% Java (+ specific or native code for system probes)  based on OW2's Fractal component model (fractal.ow2.org) © OW2 Consortium 2010 http://selfxl.gforge.inria.fr 9
  10. 10. CLIF: Self-benchmarking of JavaEE applications feedback load injector 1 probes automatic load injector 2 control probes load injector n measures Performance Response 1 JOnAS VM prediction time 2 JOnAS VM saturation 3 JOnAS VM criteria definition/choice workload © OW2 Consortium 2010 http://selfxl.gforge.inria.fr 10
  11. 11. JASMINe Virtual Machine Management (VMM)  Support for the following hypervisors:  Open-source Xen and KVM  Citrix XenServer  VMware virtualCenter 2.x  Microsoft Hyper-V © OW2 Consortium 2010 http://selfxl.gforge.inria.fr 11
  12. 12. Entropy: Architecture Entropy is a virtual  machine (VM) manager for  clusters and acts as an  infinite control loop, which  performs a globally  optimized dynamic VM  placement without  downtime according to  cluster resource usage and  scheduler objectives © OW2 Consortium 2010 http://selfxl.gforge.inria.fr 12
  13. 13. Entropy: Dynamic consolidation Our first result focuses on the dynamic consolidation issue.  Entropy observes the current CPU and memory requirements  of each VM and computes a globally optimized placement of  them that satisfy all their requirements while using a minimum  number of hosts. © OW2 Consortium 2010 http://selfxl.gforge.inria.fr 13
  14. 14. Entropy in Action Consumption is reduced by 25% 4 Tasks, 3 or 4 servers 4 tasks 4 servers © OW2 Consortium 2010 Without Entropy http://selfxl.gforge.inria.fr With Entropy 14
  15. 15. To conclude  Conclusion  Self-management of compleX and Large scale systems  Toward an Open Cloud platform  Future Work  Domain Specific Langage • Shell Console adminstration  Coordination between levels  Pointers:  ANR SelfXL project: selfxl.gforge.inria.fr  JASMINe: jasmine.ow2.org  Entropy: entropy.gforge.inria.fr  CLIF: clif.ow2.org © OW2 Consortium 2010 http://selfxl.gforge.inria.fr 15

×