Information Technology Reference
In-Depth Information
shared resource information in the distributed
computing environment. However, the above
resource monitoring systems do not support the
P2P mechanism for sharing resource information
among sites, and result in the system bottleneck
in the hierarchical architecture. Therefore, we
propose a dynamic, distributed resource monitor-
ing systems in the Uni-P2P communication model
for the P2P Grid platform to capture the dynamic
distributed resource status.
sumption that different computing nodes have the
same processing speed, memory size, and storage
space. These studies also assume that the process
of load balancing is relatively short, during which
there are neither new tasks submitted nor old
tasks finished. They claim that they could solve
the problem of archiving perfect load balance in
decentralized architecture. However, load balanc-
ing is a time-consuming process even when new
jobs are submitted or submitted jobs are migrated.
In this paper, our Uni-P2P communication model
can support dynamic job migration to balance
loads among different grid sites.
Resource Broker
The load balancing mechanisms in grid systems
can be classified into the global approach and
the local approach. The global load balancing
approach usually adopts the resource broker to
distribute load. Resource brokers consider the
usage information of grid resources, e.g., CPU
load, hard disk usage, memory load, etc., to
make decisions in order to achieve better system
performance. gLite is the middleware developed
by the E-Science project (Enabling Grids for
E-Science and Industry in Europe). The global
resource broker takes charge of distributing jobs
to different VOs in the gLite middleware. These
distributed jobs are sent to the job queues in each
VO for execution. After job submission, the global
resource broker cannot dynamically adjust the
load in each job queue, i.e., the global resource
broker does not support the function of dynamic
job migration.
In previous studies (Hu et al., 2006; Xia et al.,
2006; Xu et al., 2006), authors propose distributed
load balance mechanisms for computational grids
with the unstructured P2P architecture. In their
systems, every computing node has a job queue to
manage the job execution. These studies demon-
strate that their model always converges to a steady
load balancing state without complete knowledge
about other nodes. However, they assume that the
computational grid is a homogeneous unstructured
P2P network where computing nodes in the grids
are homogeneous. It is not practical under the as-
Process Migration
Process (or Job) migration is the action which
transfers a process between two computing nodes.
A process migration (Tanenbaum, 2007) involves
data, stacks, register contents, and the state for the
underlying operating system, such as parameters
related to process, memory, and file management
information. Process migration could improve the
load balance (Eager et al., 1986; Eager et al., 1988;
Hu et al., 2006, Iyengar et al., 2006) and the reli-
ability of distributed computing systems. Recently,
some migration technology has been raised by
adopting the checkpoint/restart technology in the
migration process. A previous study (Milojičić et
al., 2000) mentions about many process migration
algorithms in distributed computing systems. Ea-
ger copy is the simple and most common process
migration algorithm. Many previous studies (e.g.,
Lazy copy and Pre-copy) (Richmond et al., 1997)
also focus on how to enhance the effectiveness of
process migration, such as the information state
transfer, the transfer order, the process resumption,
and the network traffic reduction.
Load Balance Policy
To fully exploit the P2P Grid computing system,
load balancing is one of the key issues in achiev-
Search WWH ::
Custom Search