Market-based co-ordination in computer grids
The increasing interconnection between computers through the Internet has stimulated the creation of computational grids. Their users have access to a reliable virtual computer, which consists of many heterogeneous computing resources, including processors and hard disks. These computer resources are not visible to users, in a similar way to power grids where consumers are unaware of how electric power is generated and transmitted to the socket. Enabling the execution of computationally demanding applications, such as large- scale simulations or real-time risk analysis, computational grids are accepted to have a beneficial impact on scientific research. Nonetheless, there are still barriers preventing computational grids from reaching their full potential. To determine which computing resources are allocated to which application and scheduled at what time is one of the key issues that was addressed by the CATNETS project. Project partners suggested the introduction of economic concepts and employment of market mechanisms. Two types of interrelated markets were defined: a market for trading application services and a market for trading computational and data resources. With this distinction, a given service could be priced based on the particular resources that are made available by the hosting environment and then negotiated. The difficulty in designing such markets is that the underlying mechanism through which the participants act inevitably influences the results of trading services and resources. For instance, in a sealed bid auction the valuations of participants are not made known as in an open cry auction. Information feedback may affect the bidding behaviour of the auction participants and could therefore lead to different outcomes. The application of different market mechanisms to computational grids as an allocation and scheduling mechanism has been evaluated by researchers at the University of Karlsruhe. Mechanisms relying on the presence of centralised service and resource brokers have been compared to self-organised systems consisting of autonomous agents who follow their own interest. Detailed information on the evaluation of different approaches to allocation and scheduling computing resources is available on the CATNETS project website: http://www.catnets.org/
