Information Technology Reference
In-Depth Information
resources due to the different goals in building
different VOs.
The peer-to-peer (P2P) computing system is
another Internet-scale computing model where
computers share distributed resources via ex-
changes among the participating computers (An-
droutsellis-Theotokis et al., 2004; Li et al., 2006).
The widespread deployment of P2P computing
systems offers great potential for resource sharing.
The P2P system has the similar objective of the
grid system to coordinate large sets of distributed
resources. Therefore, many projects attempt to
integrate these two complementary technologies
to form an ideal distributed computing system
(Amoretti et al., 2005; Shan et al., 2002; Shudo
et al., 2005)
In this chapter, we propose a P2P-based mecha-
nism to form a P2P Grid platform for achieving
load balancing of distributed computing resources.
In general, the job submission in grid systems is
carried out by a global resource broker to distrib-
ute load. Here, we propose a campus-to-campus
Uni-P2P communication model to integrate the
Taiwan UniGrid (Taiwan UniGrid, 2009) and
the Taiwan TIGER system (Yang et al., 2005) by
using a P2P communication mechanism which
builds the communication pipes among sites in
different grid systems. This campus-to-campus
Uni-P2P communication model also supports a
P2P resource monitoring system that captures the
dynamic resource usage. In the P2P Grid platform,
super peers are employed to manage grid sites.
The concept of super peers, which exhibit more
powerful computing ability, bandwidth and hard-
ware capacity, is also considered in this Uni-P2P
communication model to improve the efficiency
of searching distributed resources. Moreover, we
propose a dynamic distributed load balancing
policy to improve the idle resource utilization in
the P2P Grid platform.
The rest of this chapter is structured as follows:
related works are discussed in section 2 followed
by the discussion of the system architecture in
sections 3. Experimental results are shown in
section 4. Section 5 describes conclusions and
future research directions.
RELATED WORKS
There are many middlewares (e.g., Globus Toolkit,
Unicore, gLite, etc.) which have been developed
for grid systems. Most of them focus on providing
the core middleware services for supporting the
development functionality of high-level applica-
tions. However, they usually depend on special-
ized servers to maintain the distributed resource
information. On the other hand, P2P systems
adopt decentralized resource discovery approaches
and thus do not rely on any specialized servers
to capture distributed resource information. In
this section, we present the related works of grid
information systems and load balancing policies.
Resource Monitoring Systems
There are resource monitoring software for cap-
turing the resource information, such as Ganglia,
Gstat (LCG), MDS, NWS and REMOS. Ganglia
is a distributed resource monitoring system; it
monitors system performance and system infor-
mation such as CPU load, memory usage, hard
disk usage, I/O load, and network bandwidth.
Gstat is the resource monitoring tool developed by
ASGC in order to support the members of EGEE
in handling global grid resources. Gstat supports
information such as the number of CPUs and their
load, the number of waiting/running jobs, and
the response time from GIIS. MDS (Monitor and
Discovery System) is one of the Globus Toolkits;
it supports information services and monitors/
searches grid resources. NWS (Network Weather
Services) is also a distributed resource monitoring
system. It monitors the performance of networks
and computing resources periodically, and then
predicts future system performance by real time
information. REMOS (REsource MOnitoring
System) allows the application to capture the
Search WWH ::
Custom Search