Information Technology Reference
In-Depth Information
In addition to the node information, these results will be sent to GRB from any node.
There, GRB will analyze them to select/deselect the desired nodes. We mention that
always the last collected result will be saved by GRB.
4.2 Broker Layer
In this layer we have added two new sections beside general broker's sections. The
first section is related to Request Broker section. This section must broadcast packet
to all of the nodes in grid, then it must receive and save the sent results from each
node in temporary XML database (TXD). Next, Recourse selector section will exe-
cute a Fuzzy decision Tree Algorithms on TXD (gathered result). We are doing this
task in sub-section inside Resource Selector that we call FDT executer. Whenever this
algorithm has finished its task, the next sub-section, SNJ (Selecting node for job), will
use the result of the algorithm to identify suitable nodes.
FDT Executer. This section is considered for executing FDT algorithm on TXD data.
As you know, FDT is a machine learning technique for extracting knowledge that is
nearer human decision. In this research, we have used FDT algorithm (FID3), because
it is reliable and flexible and also has a high accuracy in selecting samples. All used
samples for both training and testing are extracted from the provided database (TXD).
After that FDT algorithm was performed by FDT executer, therefore we can select a
desired class for purposed jobs. Also, Jobs can be divided in several groups: high
reliability jobs, real-time jobs, normal jobs, testing jobs and etc.
SNJ sub-layer
1. Based on the gathered results from FDT executer, this section will select appropri-
ate nodes based on job conditions. There are many parameters in this section, but
the main parameters that must be considered, are as follows:
2. Very High Reliability jobs : if we want to execute the desired job successfully with
high reliability (response time is not very important), the AHR, HRTP, ASF, HRT
measures are very important. There is a priority between these measures. For ex-
ample, to achieve high reliability, AHR and then HRTP have a high priority. Of
course, other measures are also important. SNJ will analyze these measures form
gathered results (provided by FDT executer). For example, if there are six nodes
that have almost the same AHR and HRTP, or ASF and so on, then other measures
(e.g. ART or HRT) will select to evaluate the performance of these nodes. It is pos-
sible that there are some states in that SNJ cannot select its own nodes without
limit. For example, suppose that SNJ needs to select seven nodes for doing the de-
sired tasks, and there exist only five nodes with high reliability (AHR and HRTP
over 95%) and also, if there exist other nodes with low reliability (less than 50%),
then GRB can use other parameters to decreasing risk. For example, for two re-
maining nodes, SNJ can consider HRT parameter, because this is better than other
Random-based methods. All of this will be done by SNJ. Also it can use multi-
versioning in hierarchical architecture to increase reliability [14]. In other words, it
tries to start the versions through candidate nodes in parallel and distributed form
Search WWH ::

Custom Search