**4.3 Virtual environment**

Since the code was not developed for parallel execution, it was decided to optimize resources by virtualizing nodes in order to run multiple processes on the same physical machine. In this way it was possible to create multiple virtual nodes on the same resource and increase the available nodes number, e.g., the parallelization of the system, instead of coding parallelization, to improve the overall performance. Virtualized systems also help to improve infrastructure management, allowing the use of virtual node template to create virtual nodes in a short time, speeding up the integration of new nodes on the grid and, therefore, improving the reactivity and the scalability of the infrastructure. Another advantage of virtual environment is the availability improvement, since in case of damage of a virtual node the system will be able to quickly restore it, reducing the downtime due to his recovery. The open source KVM (Hirt (2010)), has been used as hypervisor. It allows to create fully virtualized machines. The kernel component of KVM is included in mainline Linux. The basic requirements for the installation of this hypervisor is that the processor of the machine supports virtualization technology (Intel VT or AMD-V).

Fig. 7. Total execution time comparison between grid enviroment and sequential execution

experiment are shown in Table 1.

Table 2. Nodes Execution Time.

these machines three VMs have been created. To run the tests six nodes with the following configuration have been used: a Master Node, two physical Worker Nodes (wn1, wn2) and three virtualized Worker Nodes (wn3, wn4, wn5). The details of the machines used for the

Grid Infrastructure for Domain Decomposition Methods in Computational ElectroMagnetics 263

As a test-case a jet fighter aircraft (shown in Figure 1) has been discretized with a linear mesh density around 5 cm. The input file has a size of 17 MB and consists of about 156K unknowns. The plane is illuminated with a wave at the frequency of 600 MHz. The geometry has been subdivided into 67 blocks, each one described in a mesh file of about 7.5 MB. The first test performed was the execution of the entire process on a single physical machine. The node chosen for this test was the wn2: despite the smaller amount of RAM, the processing time of this machine is comparable to the average processing time of the other nodes. This node splits Node Executed blocks Number of Time [h:min:s]

master 0 0:02:55 wn1 5 10 15 20 25 30 35 40 45 50 55 60 65 13 1:42:19 wn2 1 6 11 16 21 26 31 36 41 46 51 56 61 66 14 1:22:33 wn3 2 7 12 17 22 27 32 37 42 47 52 57 62 67 14 1:24:50 wn4 3 8 13 18 23 28 33 38 43 48 53 58 63 13 1:21:38 wn5 4 9 14 19 24 29 34 39 44 49 54 59 64 13 0:56:22 wn2 all 67 6:56:51

executed blocks
