Chapter 3 2nd Part Mobile Agent Migration

1
Chapter 3 (2nd Part)Mobile Agent Migration

• Youngjung Suh

2
Contents

• 3.3 Design Issues of Agent Migration
• 3.4 Reasoning about Improved Mobility Models

3
3.3 Design Issues of Agent Migration
4
Mobility Models

• Goal of this section
• To gather information about other migration
  techniques in order to
• Discuss the design issues and design alternatives
  for agent migration, show pros and cons, and
  discover dependences between different design
  issues
• For a designer of a new mobile agent toolkit,
  because he or she must decide which migration
  technique the new system should provide.
• Develop a language to describe the migration
  technique of an existing mobile agent toolkit
• To describe the migration technique of an
  existing mobile agent toolkit in a unified way,
  which makes it easy to compare different
  approaches.

Mobility model describes almost all the important
features concerning agent migration. A mobility
model defines three views on migration issues
Users view, Agents view, Networks view
5
Mobility Models (Users view)

• The 1st view
• Is considered with all aspects of migration that
  are related to the agent programmer
• Naming and Addressing
• The agent name is very important for successful
  migration, because the constraint that no two
  agents can have the same name can be best
  validated during the migration process.
• More important with regard to agent migration is
  the structure of the agencys name. Each agency
  must have a name so that it can be identified
  whenever the agency itself or any of its
  resources must be addressed.
• Creating Agents
• When an agent is created, it must be decided
  where the agent is started first. Usually, the
  agent is started at the current agency, that is,
  the one on which the agents owner placed the
  creation command. In this case the current agency
  becomes the agents home agency.

6
Mobility Models (Users view)

• The 1st view (continued)
• Agent Code
• The other important issue that must be decided is
  the place from which code can be loaded. It is
  called a code source.
• The code source is used to load classes when the
  agent is created at its homeagency as well as
  whenever classes are not available at any remote
  agency and must be loaded dynamically on demand.
• Agent Data
• Next it must be decided what types of data the
  mobile agent can access and use. Each type of
  data has its own behavior during agent migration.
  There are four types of data that are useful for
  mobile agents
• Proxy, Static, Moving, and Copying

7
Mobility Models (Users view)

• The 1st view (continued)
• Migration
• ltMigrationgt ltInitiatorgt ltMobilitygt
  ltDestinationAddressgt lt Effectgt ltErrorgt
• Migration can be initiated by the agent itself or
  by someother instance (e.g., another agent or the
  agency)
• The type of mobility can be weak or strong, and
  in both cases further issues must be decided.
• The weakest form of mobility transmits only the
  instance variables (object state) and the code of
  the mobile agent to the destination platform.
• In a stronger form of mobility the mobile agent
  toolkit allows the programmer to define the name
  of a starting method within the go command.
• The target of a migration must be discussed. Each
  migration is directed to some target, whose
  address must be defined using a migration command
  or a ticket.
• Effect of an agent migration
• Usally, in a migration, the agent is moved
  completely to the remote agency and there is
  still only a single instance of this agent.
• Error is related to an agencys behavior if there
  is a migration error.

8
Mobility Models (Agents view)

• Lets look at the way code can be relocated
  within the network (data transmission is not an
  issue here)
• Migration Strategies
• Migration strategy is the type of code relocation
  that is used for agent migration.
• Push-all-to-next stratey
• The code of the agent (together with the code of
  all referenced objects) and the serialized agent
  are transmitted simultaneously.
• Pull strategy
• It does not transmit any code with the data
  transmission
• Pull-per-unit
• Dynamically loads code on a per-class policy
• Pull-all-units
• Loads all class files as one package immediately
  if even one class file must be loaded.
• Push-all-to-all
• As in the push-all-to-next strategy, the complete
  code of an agent is transmitted, but it is sent
  to all destination platforms the agent is going
  to visit, not only to the next destination.

9
Mobility Models (Agents view)

• Code Transfer
• Following issues concerning code relocation
• ltAgentgt ltCodeTransfergt (ltCodeCachegt)
  (ltUbiquitousClassesgt)
• (last section) code relocation strategies can be
  divided into push strategies and pull strategies.
• When the system offers a push strategy, it must
  be decided which classes must be sent to the
  agency.
• When the system offers a pull technique, it first
  must be decided what transmission unit will be
  used.
• It is important to decide whether class code is
  cached by the agency after it was loaded for the
  first time.
• An important design issue is to decide which
  classes are never transmitted, even if they are
  members of a class closure, because they are
  assumed to exist at every agency already. Those
  classes were named ubiquitous.

10
Mobility Models (Networks view)

• The last view
• Cosiders all aspects related to data
  transmission.
• The transmission strategy defines the way an
  agent is actually transmitted to the destination
  platform.

11
3.4 Reasoning about Improved Mobility Models
12
In this section

• In the previous section
• Information about current implementation
  techniques for mobile agent migration
• Reasoning about Improved Mobility Models
• Drawbacks against the background of current
  implementations and evaluate whether todays
  mobility models are able to solve any of these
  problems
• Examine factors influencing mobile agents
  performance and discuss how this performance can
  be improved.
• Investigate whether the migration strategy has an
  effect on the overall performance of mobile
  agents
• Describe the new mobility model, named Kalong.

13
Drawbacks of Simple Migration Techniques, and
Current Implementations

• The inherenct drawbacks of mobile agents are
• An agents code is typically larger than a simple
  client-server request and causes a fixed overhead
  for each imgration.
• It is only practical to send large code to remote
  agencies if server results can be decreased
  dramatically by data filtering or compression
• An agents code is transmitted to a remote
  agency, even if it is never used there.
• An agents data is transmitte as a single unit,
  which means that a mobile agent carries data
  items to all servers of the given itinerary, even
  if they are never used before reaching the home
  agency again.
• In the other case, the agent carries data items
  to several agencies, although they are never used
  at the first agencies. .This was the reason for
  poor performance in the case of a high number of
  agencies.

14
Improving the Performance of Mobile Agents

• Overview of Mobile Agents Performance Aspects
• The performance of mobile agents is influenced by
  several factors, and within the life cycle of a
  typical mobile agent,
• Several areas where performance can be improved
  (e.g., its task given by the user, the route or
  itinerary, its code size, the size of collected
  data, network parameters like bandwidth and
  latency) is found.
• To structure the discussion
• There were the following classification schema
  runtime aspects and transmission aspects

15
Improving the Performance of Mobile Agents

• Runtime Aspects techniques by which an agents
  execution time can be improved.
• Code Format
• Influences code size, and therefore, code
  execution time, and , to a lesser extent,
  transmission time.
• Machine code
• Is specific to a processor architecture family
  and cannot be executed on processors that do not
  belong to this family.
• Intermediate code
• Is the result of a compilation process that is
  performed at the agents home agency. It is a
  low-level representation, which consists of
  commands for a virtual machine.
• Code Execution
• Another aspect that influences execution
  performance is the underlying hardware
  architecture. A second aspect is the agency
  software and its optimizations on the level of
  Java code.

16
Improving the Performance of Mobile Agents

• Transmission Aspects all techniques that
  influence network load and transmission time
  during gent migration are summarezed.
• Programmers View
• The most influential factor is what level of
  mobility is supported.
• Another concern from the programmers point of
  view is the itinerary an agent has to execute.
• Agents View
• A mobile agents performance is influenced by
  code and data relocation techniques we call this
  the micro level of optimization.
• Networks View
• We see therr factors that influence mobile
  agents performance network bandwidth, network
  latency, and the overall architecture of the
  network in which the agent operates.

17
Performance and Migration Strategies

• Evaluations on the relationship between the
  performance of mobile agents and the migration
  strategies used.
• How the migration strategy influences the
  migration performance of mobile agents
• What is learned from this evaluation is
• That no migration strategy produces the shortest
  transmission time in every situation.
• In homogeneous networks ,
• Pull-per-unit is a good strategy, because code
  downloading is cheap,
• whereas in a heterogeneous network,
• code dwonloading from a faraway agency is
  expensive and should be avoided.

18
The Kalong Mobility Model

• This section introduce the new mobility model,
  named Kalong, which
• Is the synthesis of most ideas for improved
  mobility models
• The main feature of Kalong is its flexible and
  finegrained migration technique,
• Which allows an agent or its programmer to define
  new migration strategies for each individual
  migration.
• Three main aspects
• Kalong defines a new agent representation and new
  transmission units
• Kalong defines two new agency types in addition
  to the already-known home and remote agencies
  used in current mobility models
• Kalong defines a new class cache mechanism that
  prevents not only class downloading in the case
  of pull strategies but also code transmission in
  the case of push strategies.

19
Kalongs Advantages

• Using the new Kalong mobility model
• The agent has the opportunity to select classes
  that should be transmitted to the next
  destination agency,
• While other classes can be downloaded from the
  agents home agency later.
• The second advantage
• Is its ability to dynamically define code server
  and mirror agencies. Mirror agency is an
  extension of a code server agency.
• The third advantage
• Is a comprehensive technique for code caching.
  Before any class is transmitted to a destination
  agency, it is determined whether this class is
  alredy available there.

20
Summary

• Design Issues of Agent Migration
• Reasoning about Improved Mobility Models
• Proposed the new mobility model, named Kalong
• The main differences of Kalong
• Adaptive transmission of code and data
• Code server and mirror agencies
• Code caching