Distributed Dynamic Load Balancing in P2P Grid Systems - Cloud, Grid and High Performance Computing: Emerging Applications - page 291

Figure 3. System architecture of the Uni-P2P communication model

sites and the local resource information among
computing nodes. The load balance module takes
charge of load measurement, best cost site deci-
sion, job queue check and job description gen-
eration. The execution management module is
responsible for job execution. Jobs waiting in the
Condor queue can be handled in this module.
Because security is not a major concern in
P2P systems, we omit the security consideration
and instead focus on research issues about job
migration in this study.
In this chapter, the computing nodes in each site
are classified into super peers and general peers.
The general peer that starts with JXTA RDV and
Relay Service becomes a super peer, and then

the super peer starts up the resource load balance
module. The attributes and resource information
of computing nodes for general peers are sent to
the super peer. In our Uni-P2P communication
model, there are six function modules to handle
P2P communication, as shown in Figure 4.
The P2P communication mechanism first
configures the basic setting before starting the
Uni-P2P services. Then, the resource discovery
module searches and records the peer information
in the host table. The resource loading module
collects the resource usage (e.g., CPU and mem-
ory load) of general peers and records the infor-
mation in the resource table. The message receiv-
ing/sending module listens to the services at any

Figure 4. P2P communication mechanism