Information Technology Reference
In-Depth Information
In this context, several scientific projects have been leveraging the potential of
Desktop Grids. Initiatives such as Distributed.net [4], XtremWeb [17], SETI@home
[12], BOINC [2], SZTAKI [14], and OurGrid [10], among others, have shown fore-
most research results by exploiting opportunistic infrastructures mainly composed of
desktop computers. These Desktop Grid systems are based on software agents that are
installed directly on the operating system in order to manage the usage of idle compu-
ting resources.
More recently, initiatives such as VMware@Home [16], LHC@Home [8], and
UnaCloud [11], among others, are pioneers in the use of virtualization technologies to
build Desktop Grid Computing. Indeed, virtualization has appeared as an innovative
technology to enable running complete guest operating systems on top of a single
hypervisor (hypervisor type I) or a hypervisor on top of a host operating system
(hypervisor type II). This latter feature allows the deployment of single or a number
of virtual machines on off-the-shelf desktops, thus facilitating the deployment of
large-scale virtual environments aimed to support Desktop Grids. Above all, virtuali-
zation enables the deployment of customized operating systems along with the full
software stack required by scientific applications across a set of desktops that may
considerably differ in terms of hardware and/or software.
Nonetheless, when the execution of Desktop Grid systems (based on agents, virtua-
lization or hybrids) occur in parallel to tasks performed by end-users, a level of intru-
siveness is caused. Such intrusiveness is defined as the degradation in the perfor-
mance perceived by an end-user that is using a desktop while a Desktop Grid is con-
currently leveraging its idle computing resources. Since most of such desktops are
non-dedicated and temporally donated resources, the level of intrusiveness becomes a
key factor either to encourage or to dissuade the donation of idle computing resources
to support Desktop Grids for e-Science projects.
In this paper, we study how intrusiveness of Desktop Grids based on virtualization
is directly related to hardware specifications of the supporting desktops. We analyze
how technologies incorporated on several generations of modern processor architec-
tures have been consistently altering such intrusiveness. Specifically, we evaluate
how dynamic performance and energy-efficient technologies, when incorporated on
the processor of the supporting desktop, directly affect the level of intrusiveness an
end-user is able to perceive when using Desktop Grid systems based on virtualization.
The remainder of this paper is organized as follows: section 2 summarizes the re-
lated work. Section 3 presents the methodology used to conduct this research. Section
4 presents the results and discussion. Section 5 presents a set of recommendations
based on our research findings. Finally, in section 6 are presented the conclusions and
future work.
2
Related Work
In the context of benchmarking virtualization for Desktop Grids, there are two main
groups of existing work related to the research presented in this paper.
