Section: Application Domains

Application Domains

As we already mentioned, we focus on two kinds of large scale environments: clouds and peer-to-peer (P2P) systems. Although both environments have the same final objective of sharing large sets of resources, they initially emerged from different communities with different context assumptions and hence they have been designed differently. Clouds provide support for a large number of services needed by scientific communities. They usually target thousands of hosts and hundreds of users. Peer-to-peer environments address millions of hosts with hundreds of thousands of simultaneous users but they offer limited and specialized functionalities (file sharing, parallel computation).

In peer-to-peer configurations we focus on the following applications:

• Internet services such as web caches or content distribution network (CDN) which aim at reducing the access time to data shared by many users,

• Data storage of mutable data. Data storage is a classical peer-to-peer application where users can share documents (audio and video) across the Internet. A challenge for the next generation of data sharing systems is to provide update management in order to develop large cooperative applications.

• multi-player games. The recent involvement of REGAL in the PLAY ALL project gives us the opportunity to consider distributed interactive video games. Theses applications are very interesting for us since they bring new constraints, most specifically on latency.

Our second application domain is based on data sharing. Whereas most work on P2P applications focuses on write-once single-writer multiple-reader applications, we consider the (more demanding) applications that share mutable data in large-scale distributed settings. Some examples are co-operative engineering, collaborative authoring, or entreprise information libraries: for instance co-operative code development tools or decentralized wikis. Such applications involve users working from different locations and at different times, and for long durations. In such settings, each user optimistically modifies his private copy, called a replica, of a shared datum. As replicas may diverge, this poses the problem of reconciliation. Our research takes into account a number of issues not addressed by previous work, for instance respecting application semantics, high-level operations, dependence, atomicity and conflict, long session times, etc.