ISPRS Workshop on Service and Application of Spatial Data Infrastructure, XXXVI (4/W6), Oct.14-16, Hangzhou, China 
and hierarchically defined. Transparent access to heterogeneous 
hardware and software operating systems is also guaranteed. 
2. VIRTUAL ENVIRONMENTS 
2.1 Emulation vs. Simulation 
There are currently several projects aiming at providing users 
and developers with virtual computing environments. There 
exist two complementary and dual approaches, depending on 
the way existing computing resources are used. 
One approach is to emulate complex computing infrastructures 
on ad-hoc software. The second approach is to simulate simple 
environments running on complex infrastructures. 
The first approach tends to virtualize complex environments 
running on simpler infrastructures: XEN [Barham, 2003] and 
User-Mode Linux UML [UML] are examples of such projects. 
Also, VMware, a “virtual infrastructure software”, is a 
commercial product in this class [VMware], 
Another example is the XEN virtual machine monitor which 
“uses virtualization to present the illusion of (running) many 
smaller virtual machines, each running a separate operating 
system instance” (Fugure 1). This is referenced as the “emulated 
virtualization” in (XEN White Paper) and dubbed “para- 
virtualization” in (Barham;, 2003). To some extent the Linux- 
VServer [Linvserv] private virtual servers that focus on 
isolation and security for private user spaces is a similar 
approach. This is what we simply call here the emulation 
approach. 
Figure 1. The virtualization approach. 
The second approach tends to simplify the users view of 
complex environments: OpenSSI [Walker, 1999], Vgrid (vgrid, 
2003) and Kerrighed [Kerri] are such examples. They provide 
single systems images (SSI) to simulate single computing 
environments running on a set of underlying systems which are 
connected together (Figure 2). This is what we call here a 
simulation approach. 
Emulation lends itself nicely to secure and multiple isolated 
instances of (possibly heterogeneous) systems running 
concurrently on the same underlying infrastructure. It provides 
complex environments suited to the application needs, at the 
price of possibly lower performance. But theoretically, any 
complex system can be designed using this emulation approach. 
Simulation in contrast does not provide superior functionalities 
with respect to the underlying infrastructure. Its main goal is to 
mask the complexity of the underlying environments. It is 
basically made of multiple instances of (Linux) operating 
systems and computing resources (files, servers, etc) and 
provides a single interface to the users. It thus simplifies for the 
end-user access, logging, and automating execution, load 
balancing, failure recovery (by component substitution) and so 
forth. Simulation here provides superior functionalities and 
simpler interfaces to the users. 
Figure 2. The Single System Image. 
Both approaches can be considered as virtualization approaches, 
although they are very different in their goals and deployment, 
because in both cases, the users are ultimately made unaware of 
the underlying computing infrastructure. 
A direct benefit from both approaches is that various tasks can 
be automated (load-balancing, task relocation), made 
transparent (remote access to files). Further, the virtualization 
approach improves interoperability by using dedicated 
application environments, usability, security (task isolation, file 
protection) and performance (dynamic allocation of processors 
to threads). 
Another side-effect is that the underlying hardware and software 
environments are masked to the users. Consequently, various 
(heterogeneous) computing resources can be used, and their 
location is ultimately unknown. This clearly improves 
extensibility and scalability by masking the underlying 
infrastructures as well as adaptability (infrastructure changes are 
made transparent to the applications). 
Access to resources connected to a local high-speed network is 
a de facto goal for simulation environments, which clearly aim 
the cluster-computing arena (using for example cluster-wide file 
access, TCI/IP, single cluster-wide naming, etc). 
Access to local or wide-area grids can be seamlessly hooked to 
the simulated environments because dedicated computers can be 
connected which are in charge of the communication with the 
networks. 
This is where the computing grids step in. Note that they are not 
strictly required in our approach. However, the fact that they 
have long been advertised as providers of huge raw computing 
power cannot be ignored. They provide here the power to 
emulate the necessary infrastructures required by the complex 
environments supporting the applications being designed: 
spatial data infrastructures.