A Uniform Programming Abstraction for Effecting Autonomic Adaptations onto Software Systems

1
A Uniform Programming Abstraction for Effecting
Autonomic Adaptations onto Software Systems

• Valetto, Kaiser, Phung

2
Introduction of effector

• Large-scale system presents heterogeneity
• Need to achieve runtime adaptation
• Codes used to effect a desired adaptation step
  effector and generic effector
• Negative impact level of generality of autonomic
  techniques
• Best abstract away from the peculiarities of
  individual effectors or effector technologies
  with a uniform veneer, leveraged interaction with
  all effectors

3
Introduction of effector

• Abstract programming model limited set of
  primitives to describe and direct the various
  phases of work of an effector, generic effector
  API, hide diversity of effector
• All effectors can be managed uniformly through
  the effector API
• Generality in the autonomic control facilities,
  which interact with their adaptation targets via
  effectors

4
Introduction of effector

• Execute relies of Effector (some pieces of code)
  to carry out the planned changes
• Forms of effectors
• Resident algorithm
• Module inside a software component
• Proprietary operations( invoked outside the
  components)
• Functionality exposed in compliance with standard
  programming interfaces
• Mobile codes
• External utilities

5
Introduction of effector

• Adaptation made
• Turning of internal parameters
• Rearrangement of architectural configuration
• Distribution layout of a complex distributed
  system
• Controlled shut down/restart of managed resources
  (Recovery Oriented Computing)

6
Runtime adaptation with KX

• Workflakes (auto. Infrastructure KX relies on)
• Workflow place is structured as a task
  decomposition hierarchy
• Leaf task produce the intended side-effects of
  the workflow (selection/execution of a specific
  effector)
• Effectors as first class resources
• First release, mobile code, Worklets

7
Runtime adaptation with KX

• Worklet
• mobile code system,
• require at the receiving end of the code
  transfer the presence of a landing dock, which
  can receive and execute the incoming
  code--worklet virtual machine (WVM (in JVM))
• Drawbacks
• Reach of workflakes runtime adaptation was
  limited to targets into which WVM could be
  embedded
• Needed to wrap WVM with all kinds of effectors
• Or target system must be Java

8
Case study IM service

• Major adaptation targets IM servers,
  corresponding web components and BEA weblogic
  J2EE application servers
• Worklets carried classes and configuration code
  for the IM servers to various hosts in the server
  farm
• Responsiveness issues KX monitored the queue of
  pending HTTP requests, and workflakes response by
  tuning the number of threads assigned by the
  application server
• Complication (thread tuning) workflakes had to
  ship a new instance of the same mobile code
  snippet to the application server
• Implausible to select one size fits all
  effector that can be re-used efficiently across
  diverse technological settings

9
Proposed effector programming model

• Principles
• Distinction between the activation mechanisms and
  the adaptation provisions of effectors (interface
  vs. implementation and information hiding)
• Strict separation in runtime adaptation between
  the computations effecting changes and the
  adaptation logic
• Generalization of the work carried out by an
  effector according to a limited number of phases

10
Primitives presents operational semantics

• Lookup preliminary for the adaptation engine
• Recruit allows adaptation engine to get a handle
  at effectors
• Configure initialization and customization
• Activate launching the execution of the effector
  computation
• Relay provides a way for effector to report back
  to the engine any relevant data generated
• -effector catalog

11
Generic effector API

• Catalog slot always available, issue queries to
  the catalog
• Activation slot actual activation mechanism of
  an effector
• Relay slot technology-dependent plugins

12
Implementation

• Workflakes Java, Cougaar(Cognitive Agent
  Architecture Open Source Project )
• For non-Java effectors, cross-platform
  interoperability facilities by Java Native
  Interface (JNI)

13
Comments

• effector- part of KX, need understanding of KX
  to study KXs properties of autonomic computing
• achieve self-configuration and self-management
• Abstracts away from peculiarities of
  target-specific effector

14
Questions

• Load balancing of the IM system?
• Can emergence be applied?
• What if catalog response goes wrong?
• (must include some mechanism for selection)
• How to achieve activation without WVM?
• Comparing with hardwiring adaptation mechanism,
  what are the pros and cons of this method?