ns2 project in Georgia

ns2 project in Georgia

       ns2 project in Georgia although the linear mode of operation is efficient with small clusters because no overhead is imposed from the structure, with the hierarchical structure on large clusters, most agent operations are performed in parallel resulting in faster response times. The hierarchical structure also provides

other advantages such as: 1. providing scalable mechanisms to easily expand the

system, 2. providing the update and recovery ns2 project in Georgia mechanisms for

automatic detection of agent failures or change of

status/resources and techniques to report errors and

adapt to changes, 3. providing routing capabilities in the leaders to

facilitate process communications across ns2 project in Georgia multiple

platforms over multihop links, and

4. making the agents management and ns2 project in Georgia monitoring

operations more efficient and less dependant on the

full connectivity of the system.

The middleware infrastructure is capable ns2 project in Georgia of supporting

different parallel programming models. An example of this

support is the implementation of the Java Object-Passing

Interface (JOPI) [23]. In addition, distributed applications

utilize this middleware infrastructure ns2 project in Georgia to facilitate their

operation. In this section, we discuss JOPI, which provides

APIs similar to MPI and facilitates information exchange

using objects. It utilizes the features provided by the

middleware, including the scheduling ns2 project in Georgia mechanisms, remote

deployment and execution of user classes, control of user

threads and available resources, and the security mechanisms.

In addition, JOPI was designed such that processes

communicate directly with one another if all job threads are

directly connected. Otherwise, the threads utilize the

agents’ routing capabilities. Benchmark ns2 project in Georgia programs were written to evaluate the performance

of the system using JOPI. All experiments, unless

otherwise mentioned, were conducted on Sandhills, a

cluster of 24 Dual 1.2 GHz AMD-processor nodes, 256 KB

cache per processor, and 1 GB RAM per node. The cluster is

connected via a 100 Mbps Ethernet. For these experiments,

standard JVM sdk 1.3.1 was used. To test the agent overhead, Java programs were executed

independently (without the agent) and then through the

agent. The average execution time for both executions were

measured and compared. Currently, a small ns2 project in Georgia overhead

(around 0.37 percent) was found since the agent is very

lightweight. We assume that adding more functions to the

agent may introduce additional, but relatively minor delays.

In addition, the overhead is relatively ns2 project in Georgia independent from the

application, thus it will increase only as ns2 project in Georgia the number of

processors or machines used increases.