Database Reference
In-Depth Information
data model that captures the shared semantics of all the tools integrated and a bi-directional
semantic translator for each of them. This is an elegant solution to tool integration that also
avoids the scalability problem associated with pair-wise translation; i.e., the integration
of a new tool requires only a single new translator (and possibly modifi cations to existing
ones) and not N, i.e., the number of integrated tools. This is a key advantage since one of
the design goals of MILAN is to provide an open environment so that users can integrate
additional tools on their own.
The complex modeling language allows for the specifi cation of the desired application
functionality in the form of an extended datafl ow representation with strong data-typing and
parametric modeling, provides the means to specify the available hardware resources and
enables the user to defi ne mapping information between the two. Finally, application require-
ments can also be captured by explicit constraints in the models. Instead of specifying a point
solution, however, MILAN enables capturing the whole design-space of the application. At
any point in the hierarchical datafl ow, explicit design or implementation alternatives can be
specifi ed. For example, different algorithm choices optimized for speed, memory require-
ments or power consumption can be captured this way or optimized implementations can be
provided for different hardware targets. Similarly, multiple hardware resource options can
also be supplied. Finally, hardware/software allocation need not be fully specifi ed. These
techniques make it possible to describe a large — potentially exponential — set of solutions
forming the design space of the application. Our symbolic design-space exploration and
pruning technology rapidly narrows it down to the subset that satisfi es all requirements of
the system that are captured as constraints (Neema, 2001).
Tools currently integrated into the MILAN framework include such functional simula-
tors as Matlab, SystemC and ActiveHDL (a VHDL simulator) and a high-level performance
estimator, HiPerE (Mohatny & Prassana, 2002). These tools are not aware of which parts
of the system are to be implemented in hardware and which parts in software. This makes
MILAN a true system-level hardware/software co-design environment. Of course, lower-level
integrated tools, such as such cycle-accurate simulators as SimpleScalar, are already tied to
a specifi c hardware technology. Note that the different tools only communicate through the
integrated system models. This eliminates the need for each tool to be interfaced directly
to all others.
SSPF
, a predecessor of GME was used to create the Saturn Site Production Flow
(SSPF) system that monitors the car manufacturing process at GM's Saturn Corporation,
providing key production measures to managers in real-time (Long, Misra & Sztipanovits,
1998). The system models describe the manufacturing processes down to the machine level,
the buffers between the processes (e.g., conveyor belts), the instrumentation (i.e., PLCs),
and how the information is to be presented to the user. The interpreters generate various
confi guration fi les and SQL database schema to confi gure the SSPF client-server applica-
tion. The program gathers the production information, stores it in a real-time database and
makes it available to any user in the plant.
GRATIS
, a graphical development environment for TinyOS, provides an intuitive
visual interface and automatic code generation capability for the development of TinyOS-
based sensor network applications (Volgyesi & Ledeczi, 2002). With the original TinyOS
tools (Hill, 2000) working with textual confi guration fi les while developing non-trivial
applications could quickly become an error-prone and tedious process. Function-like enti-
ties have two or more names in the fi nal application; this characteristic is inherent in the
fl exible design, enabling the creation of countless different applications without touching
