GSAF: A Grid-based Services
Transfer Framework
Chunyan Miao, Wang Wei, Zhiqi
Shen, Tan Tin Wee
Motivation
• Grid provides an integrated computing
environment, facilitating maintenance and
control of information and other kinds of
resources e.g. services.
• However,
– Existing services are still tied with definite
containers.
– When new services are deployed, they come
to function only after container is restarted.
Objective
• Execute services dynamically to break the
tight coupling between services and
computers
Grid Resource Allocation
• Grid resource allocation has attracted a lot of attention in
recent years:
– Globus focuses on providing uniform and scalable
mechanisms for naming and locating computational
and communication resources on remote systems.
– GRASP [1] supports some features for user-friendly
resource allocation such as resource brokering,
scheduling, monitoring, and so forth.
– Nassif et al. [2] presented a Multi-Agent System that
chooses the best place to run a grid job by making use
of negotiation.
GSAF
• Existing work mainly focuses on how to find,
locate, select and schedule existing static
services.
• GSAF (A Grid-based Services Allocation
Framework) is proposed to:
– dynamically extend and adjust computing ability of
nodes in grid systems.
– balance the total weight by fully utilizing free or idle
computing resources.
– and provide a form of resource management to
improve the flexibility of Systems
GSAF—Resource View
• Service components viewed as explicitly
manageable resources
• GSAF partitions resources into:
– Service Components Repository (SCR): logical pool
gathering all the available service components
– Computing Nodes Repository (CNR): hardware pool
gathering available computers
– Data Sources Repository (DSR): logical pool
gathering all the data related to service components.
GSAF—R-language
• R-language: a resource-oriented workflow
description language
• Three logical elements
– Action: a definite resource processing
behavior
– Scenario: a finite series of actions
– Task: scenario which has definite and
meaningful purpose according to user
request. A task is basically a running script
GSAF Architecture
• Computers are categorized into two different
kinds of nodes:
– central nodes: responsible for central
management and scheduling such as
resource managing and task scheduling
– and computing nodes: contribute computing
ability to run services, i.e. the resources in
CNR
• Each node is controlled by an agent. The whole
system is thus a multi-agent system (MAS).
GSAF Architecture (cont’d)
• Architecture of Central Node Agent
GSAF Architecture (cont’d)
• Architecture of Computing Node Agent
GSAF Architecture (cont’d)
User Agents
Request
R-Language Generator
Computing
Node Agent 1
Task description file
R-Language Processor
Computing
Node Agent 2
launch
Service
Components
Allocation
Computing
Nodes
Management
……
Communicate
Computing
Node Agent n
……
System Components
Central Node Agent
Computing Node
Agents
GSAF Strategies
• Use service cache to deal with the service
components swapping: a distinct feature of
GSAF.
– LRU (Least Recently Used): The least recently used
service component in buffer is recorded. If
replacement is needed, swat it out.
– NRU (Not Recently Used): The service component
which hasn't been used in a certain period is recorded.
If replacement is needed, swat it out.
– FIFO (First-In First-Out): The service components are
organized in a queue according to the order of arrival.
If replacement is needed, swat out the service at the
head of queue.
GSAF—Strategies (cont’d)
• Although the best solution is to select the
most powerful computer, it may not be
practical in real use because of the
changings on-the-fly, for example the CPU
usage.
• A heuristic selection strategy is used in
GSAF, namely, weighted ranking.
Prototype
• An application of GSAF is implemented in
the field of bio data mining system.
– Use Globus Toolkit 3.2 to provide grid
environment.
– The modules of central node and computing
node are implemented as grid services in
Java supported by Globus grid service
container.
Conclusion
• GSAF is proposed to dynamically allocate
services
– Swap and execute services dynamically to break the
tight coupling between services and computers.
– All the resources are categorized and managed in
corresponding repository.
– Dynamic binding among different kinds of resources
provides a flexible pattern to execute services
• On going and Future work:
– Applications of GSAF to Bio Applications.
– Mobile Service Flow on WWW
– Trusted Service Grid
Thank You!
References
• [1] OGSA(Open Grid Services Architecture)
Documents:http://www.globus.org/ogsa
• [2] Globus: Research in Resource Management,
http://www.globus.org/research/
• [3] L. Nassif, J. M. Nogueira, M. Ahmed, R. Impey, A.
Karmouch. Agent-based Negotiation for Resource
Allocation in Grid. Workshop on Computational Grids
and Applications, 2005
• [4] R. Parra-Hernandez, D. Vanderster and N. J.
Dimopoulos. Resource Management and Knapsack
Formulations on the Grid. IEEE/ACM International
Workshop on Grid Computing (GRID'04), 2004