Information Technology Reference
In-Depth Information
Service
requestor
Define job
Main hub
CFD-
customized
service
domain
Schedule job
High
One or
more
level
Job
level
Sub hub
Sub hub
SOA
Transmit job
Web
container
GT3
Service provider
Service provider
Service provider
ADP
master
Low
Decompose sub-job
Sub-
job
level
level
Slave,
CFD
applications
Compute
node
Compute
node
Compute
node
Compute
node
Compute
node
Compute
node
Parallel
computing
FIGURE 4.3
Physical architecture of the CFD grid platform.
invocation entrance. The main hub is also responsible for the task schedul-
ing and resource allocation. It collects all the service information from sub-
hubs and then i nd the best resource solution for a given service requestor.
The subhubs also have those functions that the main hub has, but within
a smaller area. Subhubs deliver the subtasks (jobs) to the specii c service
provider, and then collect the results. All the services are provided as a
Web service. In this way (hierarchy architecture and general service inter-
face), the SOA has a good scalability.
In the low level for parallel computing, all kinds of parallel environ-
ments could be employed. In this system, MPICH2 is used as the low-layer
parallel environment. The service provider decomposes tasks that are
distributed by subhubs and then sends them to computing nodes via an
MPI mechanism. This feature enables us to generally utilize the distrib-
uted computing nodes to provide scalable and stable services.
4.4.4
Course Online Grid
Course Online Grid ( realcourse in short) is a video stream service sup-
ported by a collection of physical servers distributed all over China
[26]. Since its birth in the April of 2003, realcourse was in nonstop
 
 
Search WWH ::




Custom Search