Information Technology Reference
In-Depth Information
Fig. 3. The ratio of successfully finished jobs in both Random method and our provider
In this method we are choosing the best conditions for job. Whereas in other meth-
ods (for example, random methods), It's possible that have a high risk to select a
node. The Ratio of successful jobs in our methods is compared with another random
method [16] in below diagram (Fig 3).
In this comparison for each node, we have considered one job and the results of
testing have been repeated for many times. As you see, there are eight nodes and if
one node is selected, this best condition in our approach and with increasing the num-
ber of selecting nodes, our method will be similar to Random method. So, this dia-
gram shows that the selected nodes by our method have good performance in stable
state.
Finally, the last diagram (Fig 4) illustrates the time execution for several jobs with
different size over 4 nodes. At the first time the proposed applications is executed to
selecting best nodes. This is maybe take time less than 1 minute to a few minutes
(depend on nodes positions and network bandwidth).
As you see, when the job size and time execution is increased, the random method
has not operated as well as job that has small size an small estimation time. Because
selected nodes in random method is different from our method. On the other hand,
maybe this node for small jobs (small size and execution time) had been best effi-
ciency. Thus, it is possible that random method select this node for a desired job.
The result shows that our approach can achieve better performance under this strat-
egy. After each measure, it is seemed that, the ratio of successfully finished jobs, have
improved. It is memorable that, for all jobs smaller than 5MB and approximate time
less than 2 minutes, almost all jobs finished successfully. Therefore if there is a suit-
able method according to job requirement for selecting optimal nodes, performance of
resources will increase because we can select some nodes that they were best nodes
(in efficiency and availability) in the past. Also, time execution (for heavy processes)
and also fault rate will be decreased, because we can select some nodes that they have
much Free RAM and CPU-Idle with few processes.
Search WWH ::




Custom Search