Information Technology Reference
In-Depth Information
optimized in these projects, the migration overheads are still very high. A
more efi cient approach is the VM live migration work also from the Xen
group [10]. This method optimizes the migration of the memory state, but
its feasibility is limited to clusters with a global accessible i le system. As
all the checkpoints and resume methods, it does not address the issues of
heterogeneity and communication. The image-based migration does not
support reconi guration and does not reduce the cost of coni guration and
deployment.
A virtual cluster (V-Cluster) is a small-scale cluster emulated by the vir-
tualization technology. Projects on the V-Cluster include the Virtual Cluster
by Purdue University and the Cluster-on-Demand (COD) [22] by Duke
University. The former provides dynamic i ne-grained virtual clusters by
running unmodii ed commodity operating systems on a scalable shared-
memory multiprocessor. The latter partitions a physical cluster into mul-
tiple independent V-Clusters, where the V-Clusters are created within the
constraints of the physical cluster. These methods are designed to support
a limited coni guration management: coni guring V-Clusters on a shared-
memory multiprocessor or a physical cluster. They do not support the
modeling, incarnation, and reconi guration of a virtual environment.
In summary, most recent advances in virtualization technologies are
in extending virtual technologies to distributed and parallel environments.
There is little or no attention to system coni guration management. The
work that gives most attention to the topic has been carried out by research-
ers of Oak Ridge National Laboratory, who have developed a grid solu-
tion named OSCAR-V [23] that provides VM instantiation, management,
and monitoring tools. But their solution does not provide support for
VM migration capabilities and l exibility to work with other grid tools
(like job schedulers).
The DVM middleware is a complement of existing research and they
mutually support each other. As virtualization becomes increasingly pop-
ular, the coni guration of virtual environments becomes increasingly
timely and important. Meanwhile, a DVM solves the system coni guration
management issue via solving the coni guration of virtual environments.
It makes virtualization a better and more practical choice for the current
distributed environment. Its impact is beyond system coni guration man-
agement and extends to lift the competitiveness and applicability of the
virtualization technologies as a whole. The DVM approach is explained in
detail in the following section.
16.3 Dynamic Virtual Machine: The System Approach
A DVM middleware is proposed for coni guring virtual computing environ-
ments and managing various system coni gurations. A DVM is designed to
Search WWH ::
Custom Search