Experimental Results and Validation of DVMS
Flavien Quesnel · 2014
This chapter shows the experimental results obtained during the evaluation of Entropy and distributed virtual machine scheduler (DVMS). By analyzing the results of the simulations performed on the Grid'5000 test bed, it is observed that DVMS usually built small partitions, which allowed it to quickly find a solution to “overloaded node” and “underloaded node” problems. Moreover, the speed of DVMS had no significant negative impact on the consolidation rate, since DVMS behaved well compared to Entropy. The chapter describes that, in the case of DVMS, the time it took to apply a reconfiguration plan largely dominated the time to compute it. However, Entropy spent almost as much time computing as applying reconfiguration plans. The chapter explains that SimGrid toolkit is used to observe how Entropy and DVMS behaved while the workload injected into the VMs fluctuated while they were computing a reconfiguration plan.