Summary: | 碩士 === 國立中山大學 === 資訊工程研究所 === 83 === In heterogeneous distributed environment, load balancing is
very important for overall system performance improvement. But
load balancing is not a easy work. So in this thesis, we will
use benchmark to normalize the difference between the systems
and establish a prediction system to forecast the future load
variation by some techniques like linear regression and pattern
composition. This system will help users to decide the best
host to execute their programs for the maximum utilization of
all system resources. And it is hard to write distributed
applications in heterogeneous environment, too. For this
reason, we provide a transparency tool --- modified PVM, which
scheduler has been improved by our prediction system to
automatically balance the system load for more efficient system
performance.
|