Information and Control in Gray-Box Systems by pptfiles

VIEWS: 2 PAGES: 26

									Information and
Control in Gray-Box
Systems
Arpaci-Dusseau and Arpaci-Dusseau
SOSP 18, 2001

John Otto
Wi06 CS 395/495 Autonomic Computing Systems
Overview
n OS and Gray-Box Advantages
n Techniques
n Previous Approaches
n Case-Studies
n Gray Toolbox
n Autonomic Perspective
What is Gray-Box?
n   Premise
    ¨ Operatingsystems cannot be easily modified without
      performance risks
n   Goal
    ¨ Incorporatenew, “special application” OS ideas into
      systems without modifying the OS itself
n   Method
    ¨ Using  knowledge of OS algorithms, observe the OS
      “state” and present an optimized interface for the user
      (the Information and Control Layer, ICL)
General Capabilities
n Applications do not necessarily need to be
  designed to interface with the ICL
n Easy to port—ICLs usually assume an
  algorithm and perform general tests to
  determine the OS state.
Overview
n OS and Gray-Box Advantages
n Techniques
n Previous Approaches
n Case-Studies
n Gray Toolbox
n Autonomic Perspective
Gaining Information
n   Obtain Algorithmic Knowledge
    ¨ Trade-off   between generality and optimization
n   Monitor Outputs
    ¨ Information   in “covert channels” implies state
n   Use Statistical Methods
    ¨ Generate a “system profile” to distinguish normal and
      abnormal system performance
n   Use Microbenchmarks
    ¨ Judiciously   conduct performance tests on the system
n   Insert Probes
    ¨ Probes   help obtain, but also modify, the system state
Asserting Control
n   Exploit algorithmic knowledge to simply
    achieve a goal
    ¨ e.g.   prefetching a file
n Move the system to a known state
n Implement feedback systems
    ¨ Repeated use should optimize the ICL
    ¨ Design should keep OS in known state
Overview
n OS and Gray-Box Advantages
n Techniques
n Previous Approaches
n Case-Studies
n Gray Toolbox
n Autonomic Perspective
Existing Microbenchmarks
n Typically run in a controlled environment
n Collect static data
n Time restrictions are not imposed


n   Hence, they do not offer insight into the
    unknown state of a system—only static
    parameters
Existing Gray-Box Systems




n   Capabilities
    ¨   TCP: diagnose network congestion
    ¨   Implicit Coscheduling: run communicating processes
        concurrently
    ¨   MS Manners: optimize resource (CPU) availability for important
        processes
Overview
n OS and Gray-Box Advantages
n Techniques
n Previous Approaches
n Case-Studies
n Gray Toolbox
n Autonomic Perspective
Detailed Case Studies
File-Cache Content Detector
n   Goal
    ¨ Order data accesses to maximize cache hits,
      minimize disk accesses
n   Methods
    ¨ Internal   Simulation vs. Inference by Observation
       n   Simulation expensive, requires all processes to cooperate
    ¨ Exploit   spatial locality (page loading algorithms)
       n   Probing one region of a file can indicate whether that region
           of the file is in cache
n   Limitations
    ¨ Probing    small files significantly alters the cache state
      of that file
FCCD: Exploiting Spatial Locality
FCCD: Implementation and Interface

n   Resilient Interface
            built-in application ICL functionality
    ¨ Library:
    ¨ Command line: orders a list of files passed to
      command line tool
n   Implementation
    ¨ Differentiation    between cache hit and miss
       n   Sort files/regions of a file by shortest probe access time
    ¨ Choice of Access Unit size—minimize disk seek time
    ¨ Choice of Prediction Unit size—minimize probe use
       n   Perform a few probes per access unit (prediction unit smaller
           than access unit)
       n   Select random byte in prediction unit
FCCD: In Action
FCCD: In Action
File Layout Detector and Controller
n   Goal
    ¨   To ascertain the layout on disk of a set of files
n   “Gray-Box” Knowledge
    ¨   Most file systems localize contents of a directory on the same set of
        disk cylinders
n   Methods
    ¨   Refresh directory structure
    ¨   Use knowledge of i-node assignment to order file accesses
n   Implementation
    ¨   Call stat() on each file
    ¨   Refresh the directory
    ¨   Return list of files sorted by i-node
n   Limitations
    ¨   UNIX-oriented optimization (i-nodes!)
    ¨   Dependence of other running applications on i-node numbers
FLDC: In Action
Memory-based Admission Control
n   Goal
    ¨ Prevent   overuse of memory resources
n   Methods
    ¨ Measure   amount of memory that can be referenced
      without causing a page replacement
    ¨ Applications are notified when there is not enough
      free memory for an allocation request
n   Limitations
    ¨ Accuracy limited by page-replacement algorithm
    ¨ Just because the MAC application is “nice” doesn’t
      mean that other applications can’t cause thrashing.
MAC: In Action
Overview
n OS and Gray-Box Advantages
n Techniques
n Previous Approaches
n Case-Studies
n Gray Toolbox
n Autonomic Perspective
Gray Toolbox
n Microbenchmark results stored in common
  repository for use by ICLs at system level
n Overhead-sensitive operations use system
  -optimized “plug-in” functionality
    ¨ e.g.   timers
n   Provide tools for simple statistical
    calculations
Overview
n OS and Gray-Box Advantages
n Techniques
n Previous Approaches
n Case-Studies
n Gray Toolbox
n Autonomic Perspective?
Autonomic Perspective—Observations

n   Knowledge: In order for an autonomic tool to
    function well, the state of the system must be
    well-known.
    ¨ Hence, keeping the system in a known state is an
      important objective for autonomic tools.
n   Trust: If a system can provide evidence and
    reasons for its actions, a user is more likely to
    trust the system.
    ¨A  user interface detailing decisions and the
      benchmarks leading to an action would be beneficial.
n   Simplicity: Autonomic systems should operate
    based on known algorithms; actions would be
    predictable and explainable.
Information and
Control in Gray-Box
Systems
Arpaci-Dusseau and Arpaci-Dusseau
SOSP 18, 2001

John Otto
Wi06 CS 395/495 Autonomic Computing Systems

								
To top