Separating Cost and Capacity for Load Balancing in ABS Deployment Models

Software is often developed for a range of deployment scenarios; different versions of the software may be specialized for a number of distributed and even virtualized architectures. Since software performance can vary significantly depending on the target architecture, design decisions may need to address which features to include and what performance to expect for the different architectures. If the load of the software system depends on external parameters such as users, the software may also need to include dynamic load balancing strategies to alleviate congestion and thereby improve its own performance.

Executable models in the abstract behavioral specification language ABS can support such design decisions by explicitly modeling deployment scenarios, including load, congestion, response time, etc. This paper gives an overview of how deployment scenarios can be captured in ABS. A separation of concerns between execution cost at the object level and execution capacity at the deployment level makes it easy to compare timing and performance for different deployment scenarios early in system modeling. The language and associated simulation tool is demonstrated on an example of a virtual world framework for distributed gaming.