JiST: An Efficient Approach to simulation using Virtual Machines Rimon Barr et' al Cornell Universit

1
JiST An Efficient Approach to simulation using
Virtual MachinesRimon Barr et. alCornell
University

• Krishna Bhargava Vangapandu

2
Overview

• JiST Motivation
• JiST Design and Key points
• JiST Case Study SWANS
• Design Highlights
• Key implementation points
• Performance Notes
• Detailed performance graphs are not included

3
Some background ..

• Discrete Event Simulation
• Simulation Time vs. Real Time vs. Execution Time
• Simulation Kernel
• Eg TimeWarp OS
• Simulation Library
• Eg Compose
• Simulation Language
• Eg Simula, Parsec

4
What is JiST

• Java in Simulation Time
• Discrete Event Simulation Engine
• Executes discrete event simulations efficiently
  and transparently
• Simulation Programs are written in standard Java
  Language and runs on JVM.
• Embeds simulation semantics directly into its
  execution model
• Leverages on the advantages that the Virtual
  Machine provides

5
Motivation

• Existing simulator models
• Simulation Language
• Simulation Kernels
• Simulation Libraries
• Simulation Kernels
• Conveniently creates simulation time abstraction
• Such systems operate at process boundary, which
  is often the source of inefficiency
• Supports concurrent and distributed execution of
  simulation applications
• High costs for context switching and marshalling
  involved

6
Motivation contd.

• Simulation Libraries
• Trades off transparency for the sake of
  efficiency.
• Often ends up writing monolithic programs which
  has to embed the simulation semantics, making the
  program clutter wit lots of system calls.
• Simulation Kernel and its applications are merged
  into a single program
• High-Level optimizations are not possible
• Simulation Language
• Usually have good features for writing simulation
  programs, but often lack the features that a
  modern high level language has.
• Not many readily accept a new language
• Most of the time, lacks the portability that
  other languages offer.

7
JiST Hello World

• import jist.runtime.
• public class HelloWorld implements JistAPI.Entity
• public static void main(String args)
• System.out.println("Simulation starts ...")
• HelloWorld h new HelloWorld()
• h.myEvent()
• JistAPI.endAt(5)
• public void myEvent()
• JistAPI.sleep(1)
• myEvent()
• System.out.println("Hello World, t
  "JistAPI.getTime())

8
JiST System Architecture
Image taken from Jist.Ece.Cornell.edu
9
Some important points

• Simulation programs are executed on unmodified
  Java Virtual Machine
• VM Based simulation provides transparency of
  kernel-based approach and performance of a
  library based approach. It also provides a highly
  portable execution platform.
• All standard Java Packages can still be used in
  the simulation programs.
• JiST benefits from the automatic garbage
  collection, type-safety, reflection and several
  other properties of the Java Language. Implicit
  tight-event coupling

10
JiST Object Model

• Program state contained in objects
• All objects are contained in entities.
• Classes derived from Entity Interface
• Each entity has its own simulation time.
• Entities do not share state.

11
JiST Execution Model

• Entity methods are non-blocking. Blocking calls
  are supported through Continuations
• Entities are separated at runtime by separators
• Separators store unique entity identifier
  generated by the kernel at the time of
  initialization
• Entities communicate with each other via
  separators

12
Optimizations

• Timeless Objects
• Those objects which do not change over time.
• Zero Copy Semantics is useful to share state
  among entities
• Example A network packet when simulated, is done
  as a Timeless object. Thus every time when it is
  passed between Nodes, you need not create a new
  copy. Saves memory
• Reflection-based configuration
• Configuration files
• Script-Based Configuration
• JiST uses Reflection-based configuration.
• Also provides scripting functionality with
  BeanShell and Jython engines.

13
Rewriter Properties

• Dynamic class loader
• No source code access required
• Uses Apache BCEL
• Simulation time transformation to extend Java
  Object Model, Execution Model.
• Some of the extensions
• Timeless objects
• Reflection
• Tight Event Coupling
• Proxy Entities
• Simulation Time Concurrency

14
SWANS Introduction

• Scalable Wireless Ad-hoc Network Simulator
• Functionality similar to NS2, GloMoSim
• Built on-top of JiST as a proof of JiST concept
  and as a research tool
• Existing network simulators are not as scalable
  as SWANS.
• Simulates even 106 nodes consuming reasonable
  memory and time.

15
Design highlights

• Component-Based Architecture.
• Each component is a JiST entity.
• Simulated Network Packets are actually a chain of
  nested objects.
• Garbage Collector plays a huge role. Decides when
  to free the memory consumed by the simulated
  packets.

16
Embedding Java Based Network Applications

• SWANS can run unmodified Java based network
  applications
• Existing applications like FTP, Peer-to-Peer,
  Databases can be tested on the simulated network.
• Basic Idea At runtime, Java Sockets are replaced
  with SWANS sockets.
• AppJava is a harness that does this.

17
Signal Propagation

• To find radios within a given radius
• Naïve Signal Propagation
• Ns2, GloMoSim
• Grid-based signal propagation
• Implemented in ns2, SWANS
• Hierarchical binning
• Implemented only in SWANS

18
Performance Tests

• Compared performances of SWANS with NS2 and
  GloMoSim
• Observed that SWANS outperforms both the
  simulators in both memory utilization and CPU
  utilization
• Two tests
• Beaconing packets in NDP (Node Discovery
  Protocol)
• Zone Routing Protocol

19
Event-Throughput
20
Comparison Graphs
21
One last one .
22
JiST - PADS

• Not implemented as such
• Mechanism mentioned in the paper
• Use existing Controllers from the single-thread
  implementation and extend to a multi-threaded
  execution
• State rollback needs to be implemented for
  implementing optimistic synchronization.
• Use RMI and distribute controllers across a
  cluster of machines

23
Summary

• SWANS is designed for
• High simulation throughput
• Save memory (again GC plays a key role here)
• Run existing Java based network applications on
  the simulated network.
• This makes SWANS a great research tool.
• Wireless signal propagation has been modeled
  efficiently. Takes interference and other such
  aspects into account during simulation.
• Outperforms the existing network simulators in
  many aspects, especially scalability.

24
Advantages of JiST

• Application-oriented benefits
• Type safety, tight event coupling, debugging
• Language-oriented benefits
• reflection, object level isolation, garbage
  collection all through usage of JAVA
• System-oriented benefits
• Reduced IPC costs since the kernel and simulation
  program shares the same user space
• Kernel operates at byte-code level.
• Implicit support for extending the system to
  concurrent and distributed simulation
• Hardware-oriented benefits

25
Overview

• JiST Motivation
• JiST Design and Key points
• JiST Case Study SWANS
• Design Highlights
• Key implementation points
• Performance Notes
• Few performance graphs