Module 29 Business Layer Configuration Existing BCs and Fields
Building Self-Configuration Data Centers with Cross Layer Co-Evolution
-
Upload
paskorn -
Category
Technology
-
view
371 -
download
2
description
Transcript of Building Self-Configuration Data Centers with Cross Layer Co-Evolution
Building Self-Configuring Data Centers with Cross Layer Coevolution
Internet Data Center Simulations
Behavior PolicyEach agent/platform has its own policy for each behavior.
• A behavior policy • defines when to and how to invoke a particular behavior. • consists of factors (Fi), which evaluate environment conditions.
• Each factor is given a weight value (Wi) relative to its importance.
• A behavior is invoked if the weighted sum of its factor values exceeds a threshold.
Objectives• Make data centers (application services and middleware platforms) more autonomous, scalable, adaptable and survival to
• Improve user experience• Expand system’s operational longevity• Reduce maintenance cost
• Apply biological concepts and mechanisms• Various biological systems have achieved these requirements.
SymbioticSphere• Each application service and platform is designed as a biological entity, analogous to an individual bee in a bee colony.
Agents:• Application service is implemented as an autonomous and distributed software agent.
For example, an agent may implement a web service and contain web pages in its body.
Platforms:• A platform runs on a network host and operates agents.
Paskorn Champrasert and Junichi Suzuki University of Massachusetts, Boston http://dssg.cs.umb.edu/
Energy Exchange• Agents and platforms store and expend energy for living.
• Agents gain energy in exchange for performing their services to human users, and expend energy to use network and computing resources.
• Platforms gain energy in exchange for providing resources to agents, and evaporates energy to the network environment.
Factor ( F1)
Factor ( F2)
Factor ( Fn)
..
..
..
ThresholdInvoke behavior or not
w1
w2
w3
?ThresholdWFi
ii >∑
•A simulated network system is modeled as an Internet data center.•7x7 grid network topology.
• 49 network hosts•Each agent implements a web service in its body•There is one agent and one platform on each host at the beginning of simulation.
• 49 agents and 49 platforms
Performance RatioPerformance ratio is measured with seven performance metrics (response time, throughput, Load Balancing Index, resource efficiency, platform resource availability, agent energy level and platform energy level).
PGi denotes the performance in the metric i when agents and platforms obtain their behavior policies through evolution. Pi denotes the performance in the metric i when agents and platforms use manually-configured behavior policies.
serv
ice
requ
ests
User access point
Data Center
Host
(Simulated User)serv
ice
requ
ests
User access point
Data Center
Host
(Simulated User)
Regular Behaviors:• Each agent and platform autonomously senses its surrounding environment conditions and adaptively invokes a behavior suitable for the conditions.
For example, an agent may invoke the migration behavior to move toward a network host that receives a large number of user requests for its services.
Symbiotic Behaviors• Each symbiotic behavior is defined as a sequence of regular behaviors that an agent and its underlying platform perform in order.
• There are two types of symbiotic behaviors: agent-initiated symbiotic behaviors (A1, A2 and A3 behaviors) and platform-initiated symbiotic behaviors (P1, P2 and P3 behaviors)
Evolutionary Process• SymbioticSphere allows agents and
platforms to autonomously find appropriate values in an evolutionary manner, thereby adapting themselves to network conditions. Both regular and symbiotic behavior policies are encoded as genes of agents and platforms.
• Each gene contains one or more weight values and a threshold value for a particular behavior.
For example: A1Conditions:An agent wants to move toward a user but there is no platform running on
a neighboring host closer to the user.The agent may propose the local platform to replicate itself on the
neighboring host. If the local platform’s resource availability is low, the platform accepts the
agent’s proposal.Actions: The agent gives the platform the energy units of platform replication cost,
and the platform replicates itself on a host that the agent wants to migrate to. Results:The agent can migrate to the child platform and improve response time.
The platform can improve its health level because resource availability becomes higher.
Agents’ Regular Behaviors• Replication• Reproduction• Migration• Death
Platforms’ Regular Behaviors• Replication• Reproduction• Death
0
2000
4000
6000
8000
10000
0 2 4 6 8 10 12 14 16 18 20 22 24Serv
ice
Req
uest
Rat
e(#
of r
eque
sts /
min
)
Simulation time (hour)
Input:
This service request rate is taken from a workload trace of the 1998 Winter Olympic official website
Perf
orm
ance
ratio
Simulation Time (day)1 2 3 4 5 6 7 8 9 10
-2-1.5
-1-0.5
00.5
1 R vs RG R+S vs RG+SG
This result demonstrates that agents and platforms can successfully improve the quality of their behavior policies by themselves.
Simulation Scenarios
R: Regular Behaviors without evolutionary Process
RG: Regular Behaviors with evolutionary Process
R+S: Regular + Symbiotic Behaviorswithout evolutionary Process
RG+SG: Regular + Symbiotic Behaviors with evolutionary Process
Performance Ratio =PGi − Pi
Pi⎛
⎝ ⎜
⎞
⎠ ⎟
i=1
7
∑ (7)
Results