Partly this is achieved by formal models with predictive capabilities and partly by runtime monitoring and testing allowing us to adjust our models in an agile design-measure-evaluation cycle. The talk relates this work to eResearch services and systems supporting 'high-end' users of software across science and engineering disciplines.

 

There are open problems and attempts to generalise the theory underpinning software architecture definition and use. The presentation will look at some of these and point to possibilities arising from recent work such as the use of Galois connections. These capture a very general notion of dependency chasing that inductively follows the architectural structure of large systems for which so-called parameterised 'Galois contracts' provide the glue at connection points in the architecture. 

Galois connections are order-preserving mappings between two domains GC: A<==>B that capture complex dependency relationships.

For example if A=B is the domain of PCTL formula or equivalently certain probabilistic weighted automata models, and parameterised component interfaces are represented as such, then using the GCs, we can push assumptions about usage or deployment context through these architectural networks to discover and remove inconsistencies or strengthen assumptions and guarantees elsewhere in the architecture, always consistent with the architectural dependencies so specified. For hybrid models, using different specification domains, ie., A different from B, GC may map only some properties across between the two domain A and B. For example one of my PhD students works on hybrid architecture definitions and tool chains using Eclipse and Apache/Tuscany with links out to different model-checkers such as Prism and UPPAAL. 

 

Key to this dependency analysis is parameterisation of contracts and through them of the architectures.

 

As an example of system properties and guarantees of interest we present some recent work on timing analysis and fault-tolerance. The latter utilises specific types of generic architectures (parameterised product lines or families), in which widely distributed and parallel components are also autonomous processes or services that are coordinated in a specific way. The structure of the coordination, captured as a variation of the generic architecture, then defines both the architecture and behavioural constraints relevant for prediction.

Our measurements in our fault-tolerance testbed show that prediction with the knowledge of architectural types can be significantly more accurate and hence provide the architect with more control over design-time and run-time choices, than an architecture-agnostic approach to component-based fault-tolerance.

 

The talk will end by briefly looking at some challenges that global eResearch services face (ie., compute-, data- & software-intensive services for sciences and engineering). The hope is that this backdrop may foster research in such global services, assist in transdisciplinary engagements at each our own universities and also point at opportunities for multi-institutional and international collaboration. eResearch  has become a synonym not only for high-end computational  methods underpinning todays sciences as a 'fourth scientific paradigm' and used particularly in grid  infrastructures for astrophysics,  molecular biology or life sciences research on the human genome.  Increasingly the arts, architecture, humanities and  social sciences are transforming into 'digital humanities'  requiring computational analytics, search, tagging and digital  curation services for massively large collections of film and  audio, or crowd-sourced information gathering and interaction with thousands of volunteer citizen researchers each on narrowly defined subproblems.

SHORT BIO

Heinz is Professor of Software Engineering in the School of Computer Science and IT of RMIT in Melbourne, where he heads the Distributed Software Engineering and Architecture research group. He also holds the post of RMIT eResearch Director, heading a multi-disciplinary software support group working with researchers and IT services groups. Heinz has been adjunct professor at the Mälardalen Real-Time Research Centre at MdH, Västerås, Sweden for several years, at Monash University for a few years after moving from Monash to RMIT and has been visiting professor in a number of countries over the past 12 years.

Before joining RMIT he has held positions at Monash University, Melbourne, the Australian National University and CSIRO, Canberra, the International Computer Science Institute at the UC Berkeley, California, and the Gesellschaft for Mathematik und Datenverarbeitung, St. Augustin, (now part of the Fraunhofer). 

Heinz is an international participant in the FP7 ASCENS project and the FP7 ARTIST network of excellence.