Taking advantage of the computers connected by the Internet supplies vast computational resources to applications that can harvest them efficiently. Successful examples include cracking RSA code, SETI search for extraterrestrial intelligence and enumerating twin primes conducted at RPI. Large simulations can benefit from this computing environment. However, even more important aspect of the Internet for simulation community is wide availability of many simulation systems for modeling a particular phenomenon. These systems often assume stable or quasi-stable environment. They can be readily used to create multi-scale and multi-model simulations that are capable of representing many aspects of complex and interrelated phenomena, as long as we are able to link them correctly and efficiently.

One example when such linkage of models over the Internet is beneficial is a sea port simulation in which management of ship movements and anchoring can be best described by a discrete event simulation system. However, at different scale of details, the acceleration and velocity of each ship is affected by the weather and water currents which are best described by the atmospheric and shallow water continuous subsystems. Another example, that we recently implemented at Rensselaer Polytechnic Institute, is the spread of Lyme disease , which involves interactions of many species with large differences in their size and spatial and temporal scales of development (e.g., deers, mice and ticks). More examples include air traffic control, highway management systems, population dynamics of complex ecosystems, etc. Hence, multi-model simulation techniques are important in creating detailed models of complex, multi-scale systems and in linking existing partial models into an integrated model of more complex phenomena.

Often, the system being linked rely on different models, like discrete-event with optimistic and conservative protocols or continuous simulations based on partial differential equation representation. Such multi-model simulations require carefully defined and implemented synchronization between models. The models can be linked either by low level communication primitives (e.g., TCP/IP session) or by using mobile agents. An agent based approach lends itself to much more general and flexible linkages than those that can be supported by the low level primitives. For example, using mobile agents, different models can be linked together at run time. However, agents introduce larger communication delays than low level implementation. The very general open question is how to support efficiency of communication and synchronization in agent linked models in Web-based simulations. The more detailed open questions relevant to this challenge are discussed below.

In most cases the modeled system is large and must be parallelized to achieve acceptable execution time. The parallelism combined with explicit space representation makes the use of optimistic protocols mandatory in discrete event submodels. Hence, techniques for linking multi-mode simulation must consider the most challenging case of linking parallel discrete event simulations using optimistic protocols with a solver of partial differential equations in two or three dimensional domains. The challenging question is how an existing parallel discrete event simulation can be linked with a partial differential equation (pde) solver to form a single simulation in presence of rollback mechanism. In case of linear subsystems, a fast reverse computation algorithm can be used. However, for non-linear equations, the cost of reversing computation is larger than the cost of forward computation in the pde solver part. An important open question is how to minimize this cost.

Finally, the techniques for linking different models enable various decompositions, including a functional system decomposition in which a component models are divided into subsystems, each describing a different type of entities. The still open question is what kind of synchronization requirements such decompositions need and how their efficiency compares with the efficiency of a traditional approach.

In general, using the Internet either as a source of inexpensive computing power or as a repository of component models requires research on efficient simulation linking and synchronization over the Internet.