Symbiotic Network Architecture

1
Symbiotic Network Architecture

• DSSG
• (Paskorn Champrasert 4/27/05)

2
Content

• Introduction of Symbiotic Network
• Agent Design
• Platform Design
• Energy Exchange Design
• Symbiotic Behavior
• Simulation Plan

3
Part 0

• Introduction to Symbiotic Network Architecture

4
Symbiotic Architecture

• Design Concept
• The Symbiotic Architecture is a paradigm as well
  as middleware that enables the construction and
  deployment of network applications.
• Design Approach
• Design the networking architecture follows the
  biological principles and contains biological
  mechanisms.

5
Agents and platforms

• Modeled after biological concepts and mechanisms.
• Network application a group of agents
• c.f. Bee colony as a group of bees
• Store and expend energy for living.

Environment
Users
Service
Energy
SymbioticSphere
User request
Service
Agents
Energy
Energy
the place whereagents and platformslive as
biologicalentities.
Resource
Platform
Platform
Host
Host
Host
Energy
6
Part I

• Agent Design

7
Agents

• Agents are application components
• Each agent consists
• 1) Attributes (agent ID, age, etc.)
• 2) Body (service that the agent provides to
  users)
• 3) Behaviors
• Difference agents implement different behavior
  policies.

8
Agents attributes

• Agent ID
• Service Name
• Service Price
• Stored Energy
• Age

9
Agents Body

• The body implements a service that the agent
  provides and contains materials relevant to the
  service (e.g. application data and user
  profiles).
• For instance, an agent that implements a web
  service may contain web pages in its body.

10
Agents behaviors

• Behaviors implement non-service related actions
  that are inherent to all agents.
• Agent behaviors are the behaviors of agent which
  follow biologically-inspired behaviors.
• There are two types of agents behaviors
• agent behaviors
• symbiotic behaviors initiated by agent

11
Agents Behaviors

• Migration Agents may migrate from one platform
  to another.
• Communication. Agents may communicate with other
  agents.
• Energy exchange and storage Agents may receive
  and store energy in exchange for providing
  services to other agents. Agents also expend
  energy.
• Replication Agents may make a copy of themselves
  (replication), possibly with mutation of the
  replicas behavioral policy.
• Reproduction Two parent agents may create a
  child agent possibly with crossover and mutation
  of the childs behavioral policy.
• Death Agents also may die (death) as a result of
  lack of energy.

12
Agents Behaviors (cont.)

• Relationship maintenance Agents may establish
  and maintain relationships with other agents.
• Discovery Agents may seek for other agents of
  certain attributes by forwarding queries to
  agents that they have relationships to.
• Pheromone emission Agents may emit and leave a
  pheromone (or a trace) behind on a platform when
  they migrate to another platform.
• Environment sensing Agents may sense their local
  environment.

13
Agents Symbiotic Behaviors

• There are three symbiotic behaviors initiated by
  agent
• Preserving new generation
• (migration after replication)
• Agent Evacuation
• (migration before dieing )
• Agent Foraging
• (migration towards user )

14
Symbiotic Behavior initiated by agent Preserving
new generation
Motivation 1 Agent wants to replicate 2 Platform
has - Low Energy Level - Low Healthy Level
Benefit 1. Agent can migrate toward to platform
that work on healthier host. 3. Platform can
replicate
Agent wants to replicate but the
platform that it works on has low healthy level
and low energy level. Then, agent pays some
energy for its replication and extra energy for
letting platform can replicate. If platform can
replicate agent and its child can migrate to
healthier platform.
15
Symbiotic Behavior initiated by agent Agent
Evacuation
Agent is dying
migrate
Motivation 1 Agent is going to die 2 Platform
has - High Energy Level - Low Healthy
Level 3. There is healthier platform
exists.
A
A
3
Energy for Migration
Energy for letting agent migrate
1
2
Platform
Benefit 1. Agent can migrate toward to platform
that works on healthier host. 3. Platform
increases its healthy level.
Platform
High Energy Level Low Healthy Level
Platform on healthier host
There is healthier platform exists in the
netsphere but agent can not migrate to because it
has very low energy and it is going to die. Then
platform, that this agent is working on, gives
some energy to the agent to let it migrate to
healthier platform.
This scenario make weakest platform die first
16
Symbiotic Behavior initiated by agent Agent
Foraging
Agent want to migrate
migrate
Motivation Agent wants to migrate to host that
close to user but there is no platform on that
host.
A
A
5
Extra Energy for letting platform replicate
Energy for Migration
destination host address information.
1
2
4
Platform
Benefit 1. Agent can migrate toward to user 3.
Platform can replicate.
replicate
Platform
3
Replicated platform on destination host
Agent wants to work on host that close to
user but there is no platform on that host. Then
agent gives destination host address and some
energy to its platform. Platform can replicate to
destination host and agent can migrate to the
replicated platform.
17
Part II

• Platform Design

18
Platform

• Each Platform consists
• 1) Attributes (Platform ID, healthy
  level.)
• 2) Runtime Services
• 3) Behaviors

19
Platforms attributes

• Platform ID
• Stored Energy
• Healthy level table

20
Healthy Level Table

• Healthy level is property of a host, calculated
  as a function of
• Resource availability
• e.g. CPU cycles, memory space and network
  bandwidth
• Age how long a host is alive (how a host is
  stable).
• Healthy level table is a set of healthy level
  data obtained from neighboring hosts. The values
  in the healthy table are the compared values of
  the neighboring host healthy level and the local
  host healthy level.

21
Runtime Services

• Platform keeps executable of behaviors of agents.
  
• The agents behaviors (replication, migration,
  death,) will be kept as runtime services in
  platform.

22
Platforms behaviors

• Behaviors implement non-service related actions
  that are inherent to all platforms.
• Platform behaviors are the behaviors of platform
  which follow biologically-inspired behaviors.
• There are two types of platforms behaviors
• platform behaviors
• symbiotic behaviors initiated by platform

23
Platform Behaviors

• Replication Platforms may make a copy of
  themselves when their energy levels are high and
  they find healthier platforms.
• Death Platforms may die due to the lack of
  energy. A dying platform will uninstall itself
  and release resources it maintains.
• Healthy level management
• Maintaining local hosts healthy level Platform
  may update its healthy level automatically to
  indicate its age and resource availability.
• Inquiring neighboring hosts healthy level
  Platform may inquiry neighboring hosts healthy
  level to compared with its healthy level and
  maintain the healthy level table.
• Responding to inquiries from neighboring
  platform Platform may response to inquiries from
  neighboring platform by sending its healthy level
  to.

24
Platforms Symbiotic Behaviors

• There are three symbiotic behaviors initiated by
  platform
• Platform transferring
• (Agent migration after platform replication )
• Increasing Healthy level
• Platform Evacuation
• (Replication before dieing )

25
Symbiotic Behavior initiate by Platform Platform
transferring
migrate
Motivation 1. Platform want to replicate 2.
Platform has - Low Healthy Level -
High Energy Level
A
A
4
2
Energyto let agent migrate to replicatedplatform
Energy for Migration
3
Platform
Benefit 1. Agent can migrate toward to platform
that work on healthier host. 2. Platform
increases its healthy level.
replicate
Platform
1
High Energy Level Low Healthy Level
Replicated platform on healthier host
After platform replicated to healthier
host, platform gives some energy to agent to let
agent can migrate to replicated platform.
This scenario allows migration of platform
26
Symbiotic Behavior initiate by Platform
Increasing Healthy level
migrate
Motivation 1. Platform has - Low Healthy
Level - High Energy Level 2. There is
platform that works on healthier host exists.
A
A
3
1
Energyto let agent migrate to replicatedplatform
Energy for Migration
2
Platform
Benefit 1. Agent can migrate toward to platform
that work on healthier host. 2. Platform
increases its healthy level.
Platform
High Energy Level Low Healthy Level
Platform on healthier host
Platform has low healthy level and high
energy level. There is platform that works on
healthier platform exits. Then platform gives
agent some energy to let agent migrate to the
other platform.
27
Symbiotic Behavior initiate by Platform Platform
Evacuation
migrate
Motivation 1. Platform is dying and has -
Low Healthy Level - Low Energy Level 2.
There is no platform that works on healthier host
exists.
A
A
4
1
Energyto let platform replicate to healthier host
Energy for Migration
3
Platform
Benefit 1. Agent can migrate toward to platform
that work on healthier host. 2. Platform
increases its healthy level.
replicate
Platform
2
Replicated platform on healthier host
Low Energy Level Low Healthy Level
Platform is dying
Platform has low healthy level and very low
energy level. It is going to die and there is no
other platform in the netsphere. Agent gives
platform some energy to let platform can
replicate to healthier host. If platform can
replicate agents can migrate to the replicated
platform.
28
Part III

• Energy Exchange Design

29
Introduction

• In the symbiotic architecture there are two
  biological entities which are agent and platform.
  
• Agents and platforms live on energy Agent give
  the energy that it get from the user to platform
  to trade with the resource. Platform pays energy
  to environment for living.
• Energy is used to adaptively balance the number
  of agents and platforms (reflecting changing
  network environments and user requests) to
  optimize the resource efficiency in the entire
  system.
• We have to find the mechanism of paying energy
  from agent to platform, and platform to
  environment (host)

30
Energy Exchange Model
Eua Energy transferred from user to agent EPap
Energy periodically transferred from agent to
platform EPph Energy periodically transferred
from platform to host (environment) EP Energy
that agent and platform pay periodically. The
rate of EP have to be changed follow the Eua
rate.
31
Some support idea about energy transferring in
the food chain

• Biology concept 10 Rule
• (from website)
• In a food chain, an animal passes on only about
  10 percent of the energy it receives. The rest is
  used up in maintaining it's body, or in movement,
  or it escapes as heat. The amount of available
  energy decreases at every trophic level, and each
  level supports fewer individuals than the one
  before..

100 of Energy
The number in the picture is dry weight units
(g) Biomass Estimate dry weight of the
population in the area
32
Adaptive Respiration Rate

• The graph below show the respiration rate of a
  humming bird at different times of the day and
  the night (shaded).

http//www.saburchill.com/ans02/chapters/chap034c.
html
33
Finding EP
From Biology 10 Rule (the amount of energy
transferred to next level is only 10
) Therefore EPap 0.1 Eua EPph 0.1 EPap
The problem is what is the rate of energy loss ?
(frequency?)
34
Energy loss rate (F)

• Fua user request rate
• Fap energy loss rate from agent to platform
• Fph energy loss rate from platform to host
• Condition
• If user request rate is high (Fua high)
• Energy loss rate is high (Fap and Fph high)
• If user request rate is low (Fua low)
• Energy loss rate is low (Fap and Fph low)

35
Goal

• Need an autonomous mechanism to adjust the
  frequency of energy payment (Fap, Fph) depending
  on a given Fua to optimize resource efficiency of
  the entire system.

36
Approach

• Agent and Platform as Biological Entities
• User gives energy to agent when user request
  service. It is equal to agent get food from the
  user. Energy food
• Agent and Platform which behave as biological
  entities should learn something in the past to do
  their behavior for the future.

37
Approach

• In the real situation, agent cannot know the user
  request rate (frequency) but it can keep data of
  interval time (waiting time between two user
  requests)
• In order to change Fap in the future, agent need
  to keep Fua in the past
• To statistically represent the Fua in the past
  these values are used
• Average Interval Time (ta)
• Shortest Time Interval Time (ts)
• Longest Interval Time (tl)

38
Energy Exchange Algorithm

• Agent keeps the interval time between user
  request
• Agent updates shortest time, average time,
  longest time after getting user request by
  compare the new interval time and the previous
  interval time
• Agent pay energy after shortest interval time,
  average interval time and then longest interval
  time until the new user request arrives.

39
Energy Exchange Algorithm (cont)

• Because agent pay energy 10 of energy that it
  gets. (if user pays 10 energy unit per request
  agent will pay 1 energy unit to platform)
• the shortest interval time, average interval
  time, and longest interval time will be divided
  by 10 to make the interval time in the same
  scale.
• For example if the user request average
  interval time 3 minutes. Agent will loss all
  energy after 30 minutes (310) which is too large
  number.Therefore after average interval time is
  divided by 10. Agent will loss all energy after
  3 minutes (0.310) which fits to the user request
  rate.
• Therefore
• The interval time that agent pays the energy
  (shortest, average and longest) will divided by
  10.

40
coefficiency a

• There is some co-relation between the frequency
  of user request receipt by an agent and the
  agents energy payment frequency.
• coefficiency a indicates the level of
  co-relation
• The interval of the agents energy payment
• a the interval of user request receipt by an
  agent
• We use a 0.10 here

41
Window Size

• After long running the simulation, the shortest
  interval time will be very small number and
  longest interval time will be very large number.
• Therefore shortest interval time and longest
  interval time will be reset after some period.
• Assume 20 user requests can represent user
  request rate. After 20 user requests, the
  shortest interval time and longest interval time
  will be reset.
• Window size 20

42
Example

• Input
• First user request interval 3 minutes
• Second user request interval 4 minutes
• Third user request interval 5 minutes
• Ouput
• Time Energy Total Stored Energy state Shortest
  Time AverageTime LongestTime
• 3 Get 10 - 3 3 3
• 3.3 Loss 9
  Shortest 3
  3
  3
• 3.6 Loss 8 Average
  3 3 3
• 3.9 Loss 7 Longest
  3 3 3
• 4 Get 17 - 1 2 3
• 4.1 Loss 16 Shortest 1 2 3
• 4.3 Loss 15 Average 1 2 3
• 4.65 Loss 14 Longest 1 2 3
• 4.75 Loss 13 Shortest 1 2 3
• 4.95 Loss 12 Average 1 2 3
• 5 Get 22 - 1 1.6 3
• 5.3 Loss 21 Longest 1 1.6 3
• 5.4 Loss 20 Shortest 1 1.6 3

43
Simulation Model for Energy Exchange Algorithm

• Input
• User request Interval time is randomly changed
• base /- randomNumber (0-3)
• Base increase or decrease 2 unit after 50 user
  requests are generated.
• Initial Energy for Agent 100
• User give energy to agent 10 unit for each
  request.
• Agent pays energy 1 unit to platform (10 law)
• Parameter
• Window size 20
• coefficiency a 0.10

44

Model 1, user request interval time is decrease,
user request rate is increase
Input User Request
User request interval time (minutes)
User request th
There are 400 user requests in the model
45
Result Evaporate Interval Time
Model 1, user request rate is decreased, the
evaporate interval time is decreased
Evaporate interval time (minutes)
Evaporate th
Average Time
Shortest Longest Time
46
Result Stored Energy at agent
X Energy Unit of the agent (only 1 agent) Y
user request (th)
47
Model 1 Result

• When user request rate increases
• Energy loss rate (evaporation rate)
  increases.
• Agent will increase its stored energy.

48
Input User Request
Model 2, user request rate is decreased
User request interval time (minutes)
User request th
There are 400 user requests in the model
49
Result Evaporate Interval Time
Model 1, user request rate is decreased, the
evaporate interval time is decreased
Evaporate interval time (minutes)
Evaporate th
Average Time
Shortest Longest Time
50
Result Stored Energy at agent
Finally, agent die
X Energy Unit of the agent (only 1 agent) Y
user request (th)
51
Model 2 Result

• When user request rate decreases
• Energy loss rate (evaporation rate)
  decreases.
• Agent will loss all its stored energy (die) after
  user request rate decreases for a while.

52
Change Coefficient a
X Energy Unit of the agent (only 1 agent) Y
user request (th)
Window size 20
53
Change Window Size
X Energy Unit of the agent (only 1 agent) Y
user request (th)
Coefficient 0.11
54
Model 3

• The user request interval time is unchanged

55
Input Interval of user request
Model 3, user request rate is unchanged
Y interval time (minutes) X user request (th)
56
Output Interval of Evaporate Time
Shortest Longest Time
Y interval time (minutes) X evaporate (th)
Average Time
57
Output Stored Energy of agent
Stored Energy is nearly closed to 100 (initial
energy unit)
X Energy Unit of the agent (only 1 agent) Y
user request (th)
58
Model 4

• The user request interval time is dynamically
  changed

59
Input Interval of user request
Model 4, user request rate is extremely changed
Y interval time (minutes) X user request (th)
60
Output Interval of Evaporate Time
Y interval time (minutes) X evaporate (th)
Shortest Longest Time
Window size for Average Time 400
Average Time
61
Output Stored Energy of agent
X Energy Unit of the agent (only 1 agent) Y
user request (th)
Finally, Agent die because user request rate is
very low
62
Output Interval of Evaporate Time
Y interval time (minutes) X evaporate (th)
Shortest Longest Time
Update Average Time when It is too different from
shortestlongestt
Average Time
63
(No Transcript)
64
Summary

• So the energy exchange algorithm works well for
  changing the energy loss rate to user request
  rate.
• We can change coefficient a to let agent die
  faster.
• Agent and Platform will be implemented the energy
  exchange algorithm to let them loss energy.
  Finally, when the user request rate is very low
  the agent will die before platform.

65
Part IV

• Simulation Plan

66
Simulation Goals

• To show platforms achieve autonomous resource
  adaptation
• Autonomous resource adaptation
• An ability of platforms to autonomously adjust
  resource availability and resource efficiency for
  agents
• Resource availability
• How much resources are available for agents?
• Resource efficiency
• How efficiently do platforms use resources?

67
Simulation Environment

• Hosts
• 4x4 (16) hosts
• Grid topology
• Each host can run a platform.
• Each host provides three types of resources for a
  platform.
• CPU
• Memory
• Network bandwidth
• Each platform
• consumes a certain amount of resources provided
  by an underlying host.
• maintains and provides the remaining resources to
  agents.

68

• Homogeneous and heterogeneous networks
• Homogeneous
• Difference hosts have the same amount of
  resources available.
• Heterogeneous
• Different hosts have different amount of
  resources available.
• Static and dynamic networks
• Static
• The number of hosts does not change dynamically.
• Dynamic
• The number of hosts dynamically changes.
• Hosts can dynamically enter and leave the
  network.

69
Simulation Configurations

• Configuration 1
• Homogeneous and static network
• Platforms as non-biological entities.
• Only one platform on 16 hosts.
• Configuration 2
• Homogeneous and static network
• Platforms as non-biological entities
• 16 platforms on 16 hosts (a platform on each
  host).
• Configuration 3
• Homogeneous and static network
• Platforms as biological entities.
• 16 platforms on 16 hosts (a platform on each
  host)
• Configuration 4
• Heterogeneous and static network
• Platforms hosts as non-biological entities.
• 16 platforms on 16 hosts (a platform on each
  host)
• Configuration 5
• Heterogeneous and static network
• Platforms as biological entities.

70
Simulation EvaluationResource Availability (1)

• Run simulation configurations 1 and 3.
• Expected results
• In config 1, the number of platforms does not
  increase, even if user request rate becomes high.
  
• In config 3, the number of platforms can
  dynamically increase, if user request rate
  becomes high.
• Key measures
• The number of platforms
• The total amount of resources available for
  agents
• Key observation we will make
• Platforms dynamically adjust resource
  availability depending on user request rate and
  the number of agents

71
Expected Results (new added)
72
Expected Results
User request rate
Input
time
Outputconfig 3
Output config 1
of platform
of platform
1
1
time
time
73
Simulation EvaluationResource Availability (2)

• Run simulation configurations 2 and 3.
• Expected results
• In config 2, the number of platforms does not
  decrease, even if user request rate becomes low.
• In config 3, the number of platforms can
  dynamically decrease, if user request rate
  becomes low.
• Key measure
• The number of platforms
• The total amount of resources available for
  agents
• Key observation we need to make
• Platforms dynamically adjust resource
  availability depending on user request rate and
  the number of agents

74
Expected Results
Output config 2
Output config 3
of agent
of agent
time
time
75
Expected Results
Outputconfig 3
Output config 2
of platform
16
time
76
Simulation EvaluationResource Efficiency (1)

• Run simulation configurations 2 and 3.
• Expected results
• In config 2, the total resource utilization does
  not decrease, even if resource that users require
  is less than the preempted resource. There is
  unutilized resource.
• In config 3, the total resource utilization can
  dynamically change to fit with the resource that
  users require.
• Key measures
• The total amount of resources consumed by agents
  and platforms
• The number of throughput/the total amount of
  resources consumed by agents and platforms
• Key observation we will make
• Platforms dynamically adjust resource utilization
  depending on user request rate and resource
  availability on hosts.

77
Expected Results
Input
User request rate
time
Outputconfig 3
Outputconfig 2
of Agent
of Agent
20
20
time
time
78
Expected Results
Input
User request rate
time
Outputconfig 3
Output config 2
of Platform
of Platform
16
10
5
time
time
79
Expected Results
Input
User request rate
time
Outputconfig 3
Total Resource utilization
Output config 2
Total Resource utilization
160
160
time
time
80
Expected Results
Input
User request rate
time
Outputconfig 3
Throughput/resource utilization
Output config 2
Throughput/resource utilization
80
10
time
time
81
Simulation Evaluation Resource Efficiency (2)

• Run simulation configurations 3 and 4.
• Expected results
• In config 3, the total resource utilization can
  dynamically change to fit with the resource that
  users require..
• In config 5, the total resource utilization can
  dynamically change to fit with the resource that
  users require. If there are some hosts that have
  high resource availability, platform will
  replicate to those hosts. Total resource
  utilization of platform and agents is decreased
  but resource availability to agents is
  maintained.
• Key measures
• The total amount of resources consumed by agents
  and platforms
• The total amount of resource availability for
  agents/ of platform
• The number of throughput/the total amount of
  resources consumed by agents and platforms
• Key observation we will make
• Platforms dynamically adjust resource utilization
  depending on user request rate and resource
  availability on hosts.

82
Expected Results
Input
User request rate
time
Config 3 homogeneous
Config 5 heterogeneous
Assume each host has 50resource units.
Assume host has 50 resource units. host has 80
resource units.
83
If we run the simulation for long time

• In config 3, the number of platforms is three.
  The total amount of resources available for
  agents is 150.
• In config 5, the number of platforms is only two.
  The total amount of resources available for
  agents is 160.
• Using the environment sensing behavior, each
  platform can sense the resource availability on
  its neighboring hosts. During a replication
  process, it replicates itself to the neighboring
  host whose resource availability is higher than
  the local host and higher than any other
  neighboring hosts. Platforms tend to reside on
  resource-rich hosts.

Config 3 homogeneous
Config 5 heterogeneous
Assume each host has 50resource units.
Assume host has 50 resource units. host has 80
resource units.
platform
84
Input
User request rate
time
Resource Available/platform
Output config 3
Resource Available/platform
Output config 5
80
50
time
time
150/3
160/2
85
Expected Results
Input
User request rate
time
Outputconfig 5
Throughput/resource utilization
Output config 3
Throughput/resource utilization
40
10
time
time