The memory unit used in this paper has the maximum capacity and the memory VX-11e are arranged based on the adopted frequency of corresponding rescheduling strategies stored in the memory cells. When a new memory cell appears, it is added to memory unit if the maximum capacity is not overflowed. Otherwise, the cell with the least frequency would be abandoned.
Due to this memory mechanism, if the same resource fault happens again, the IRW can give a quick response by the corresponding rescheduling strategy with memory cells, which is similar to the case-based reasoning (CBR) process [44] and [45].
4. Performance evaluation
4.1. Experimental setup
In this section, we describe the overall setup of our experiments to evaluate the proposed IRW. Experiments are conducted with the WorkflowSim toolkit [46], which is a modern simulation framework aimed for workflow scheduling in Cloud computing environments. The hosts in simulation testbed are heterogeneous and modeled as performance equivalent to 1000, 1500, 2000 or 3000 MIPS. Each host can be virtualized to multiple VMs. The performance of CPU (MIPS), memory capacity and storage capacity of these VMs are different with each other and designed randomly in certain ranges. The minimum fault rate of VMs is 5% and the maximum is 15%. We also design region of division the VMs in resources pool are divided into 3 clusters by k-means in the resources manager. The relevant parameters about the experiments are presented in Table 1.
Due to this memory mechanism, if the same resource fault happens again, the IRW can give a quick response by the corresponding rescheduling strategy with memory cells, which is similar to the case-based reasoning (CBR) process [44] and [45].
4. Performance evaluation
4.1. Experimental setup
In this section, we describe the overall setup of our experiments to evaluate the proposed IRW. Experiments are conducted with the WorkflowSim toolkit [46], which is a modern simulation framework aimed for workflow scheduling in Cloud computing environments. The hosts in simulation testbed are heterogeneous and modeled as performance equivalent to 1000, 1500, 2000 or 3000 MIPS. Each host can be virtualized to multiple VMs. The performance of CPU (MIPS), memory capacity and storage capacity of these VMs are different with each other and designed randomly in certain ranges. The minimum fault rate of VMs is 5% and the maximum is 15%. We also design region of division the VMs in resources pool are divided into 3 clusters by k-means in the resources manager. The relevant parameters about the experiments are presented in Table 1.