Peer to peer computing

The use of computer networks to solve computationally demanding tasks in general has been investigated broadly for several decades now. Diverse software solutions are already available from different communities. Among them, the most famous are the volunteer based ones, especially the BOINC infrastructure. A different platform is the ACT-DC [1], a general purpose distributed computing environment developed for use in the internal network of the European Space Agency and used experimentally in the past for relatively small tasks such as ionospheric data processing, nanostructured material design and spacecraft trajectory optimisation.

In most of the implementations of distributed computing environments, the different CPUs are coordinated in a centralized (master-slave) fashion while the task gets distributed by the central server according to a selected strategy. This type of network topology ("star"-like) is, though, not the only possible one nor necessarily the best. It seems that there are no scientific publications on the effect of the network topology on the efficiency of different global optimisation algorithms. Besides, the number of different networks configurations proposed is constantly increasing. They are used for different purposes by peer-to-peer (P2P) systems [2]. These systems are fully decentralized, highly autonomous and easily scalable. The decentralization makes the network fault tolerant, efficient and sometimes cheap (since there is no need to maintain a server).

In the same direction, the gossip communication model [3], is a way to broadcast messages within a network in an asynchronous fashion. The main properties of this approach are: democratic, scalable, robust and reliable. There are several ways to implement this kind of communication: In the push-pull model the active thread initiates communication and receives peer state. The passive thread mirrors this behavior. The active nodes randomly select some other nodes (this is an important component of the model, determines the performance and the reliability of the protocol) to exchange data. The overlay topology (i.e. "who is connected to whom") of this network could dynamically change over time. This has a major impact on many functions that can be used in P2P networks, e.g. load balancing [4], data aggregation [5] and global optimisation [6], just to mention some. A P2P implementation of a global optimisation algorithm (namely the Particle Swarm Optimisation) using gossip communication model has already been published [6] with promising results.

References

1. M. Cs. Markot. 2007. "The ACT Distributed Computing Environment v.2.3 administrators guide and instructions for further development". Special publication of the European Space Agency, STR253.
2. Milojicic, D. S., V. Kalogeraki, R. Lukose, K. Nagaraja, J. Pruyne, B. Richard, S. Rollins, and Z. Xu. 2002. "Peer-to-peer computing". Technical Report HPL-2002-57, HP Lab.
3. Demers, A., D. Greene, C. Hauser, W. Irish, J. Larson, S. Shenker, H. Sturgis, D. Swinehart and Doug Terry. 1987. "Epidemic algorithms for replicated database maintenance". In Proceedings of the 6th Annual ACM Symposium on Principles of Distributed Computing (PODC'87), pages 1–12, Vancouver, British Columbia, Canada, ACM Press.
4. Jelasity, M., A. Montresor, and O. Babaoglu. 2004. "A modular paradigm for building self-organizing peer-to-peer applications". In G. Di Marzo Serugendo, A. Karageorgos, O. F. Rana, and F. Zambonelli, editors, Engineering Self-Organising Systems, number 2977 in LNCS, pp 265-282. Springer.
5. R. van Renesse. 2003. "The Importance of Aggregation. Future Directions in Distributed Computing", Springer. Lect. Notes Comput. Sci. 2584, pp.87-92.
6. Brunato. M., R. Battiti and A. Montresor. 2007. "GOSH! Gossiping Optimization Search Heuristics", In Learning and Intelligent Optimization Workshop LION2007, Andalo, February 12-17.

Outcome