Information Technology Reference
In-Depth Information
order to avoid bias produced by incidental or cumulative consumption of computing
resources.
3.3
Benchmark Selection
In order to represent and assess the intrusiveness in the execution of virtual environ-
ments for Desktop Grids, two benchmarks were set and performed over the selection
of processors. Both benchmarks were exclusively based on CPU intensive workloads
generated through the execution of a program able to calculate prime numbers by
using the sieve of Eratosthenes [9]. This integer arithmetic algorithm was selected in
order to avoid potential bias caused by improvements on the architectures tested to
perform floating-point operations. Moreover, the execution of this program had neg-
ligible demands of RAM memory and no network activity was required.
The first benchmark was designed to test the virtual environment. For this purpose,
the program was set to run on the virtual machine for approximately 10 hours thus
loading permanently it with CPU intensive tasks. As explained formerly, according to
the test scenario, the number of virtual machines varied from zero to the number of
physical CPU cores.
The second benchmark was designed to test the physical environment. For this end,
a number of concurrent processes were executed simulating an end-user without root
privileges. According to the test scenario, this number ranged from 1 to the number of
CPU threads. The concurrent processes were the same CPU intensive workloads used
in the virtual environment. However, in this case the program was set to run for only
60 seconds on average.
Each test scenario consisted of the simultaneous execution of both benchmarks.
The measurements were obtained only from the second benchmark. While the first
benchmark was executed once to permanently load the virtual machine, the second
was executed 24 separate times to obtain averaged results. This number allowed dis-
carding the two measurements that were most distant from the median in order to
avoid biases mainly produced by periodical processes running at the host operating
system level.
3.4
Intrusiveness Definition
In the context of this research, intrusiveness is defined as degradation in the perfor-
mance perceived by an end-user that is using a desktop while it is opportunistically
utilized by a Desktop Grid system based on virtualization. Such intrusiveness is ex-
clusively measured in terms of the performance delivered by the processor thus dis-
carding intensive operations related to RAM memory, storage, network or any other
computing resource. It is calculated as a percentage, named intrusiveness percentage
(
%
), and its formula is explained as follows:
%100
(1)
