ABSTRACT
Desktop grids (DGs) are grid systems where the resources are collected typically from existing desktop machines via cycle scavenging. The main advantage of DG systems compared to the other forms of grid systems, called service grids (SGs) is that they do not need a large investment and yet they can collect large resource capacity. DGs are also much simpler than SGs concerning the software infrastructure they are relying on. Installation and maintenance of DG resources is extremely simple, requiring no special expertise, so even home computers can easily be provided as computing resources. Typical DG systems are BOINC [And04], Condor [TTL04], OurGrid [CBA+06], XtremWeb [FGNC01], SZDG (SZTAKI Desktop Grid) [KKF+]. These were characterized and compared in detail in [KKF+]. One form of DG system is called volunteer computing when the resources are provided in a volunteer way, typically by home computers. In order to attract home computers, some form of reward mechanism should be introduced in the DG system. BOINC is by far the most advanced volunteer desktop grid middleware, providing credit collecting, administrating, and displaying functionalities for the project participants who donate the spare cycles of their desktop computers. The other important requirement of volunteer computing middleware is to provide some form of protection against erroneous and malicious donated computers. In this respect, again BOINC is the most advanced. It supports redundant computing and result validation. A very important aspect of volunteer computing is its community-based nature. BOINC provides a middleware framework based on which scientific projects like SETI@home [ACK+02], Einstein@Home [A+09], Folding@Home [BEJ+], etc. can be created. However, this framework should be filled in by desktop resources via the activity of a community. This community is the volunteers who want to support the given project due to their personal motivations for supporting science or other reasons. One of the decisive key factors in the success of different volunteer computing projects is their capability of attracting donors to support them. Successful volunteer computing projects are typically able to collect desktop resources in the range of 10, 000 − 1, 000, 000 that is, much bigger than the typical number of resources collected in SG systems. This shows the power of communities. Of course, these volunteer resources are not as reliable as the managed clusters
and supercomputers typically used in SG systems and hence as it was mentioned above this is an important research area of volunteer DG systems: how to handle and manage the unreliable resources.
