A peer-to-peer architecture for data-intensive cycle sharing
Proceedings of the first international workshop on Network-aware data management
Future Generation Computer Systems
A proximity-aware load balancing in peer-to-peer-based volunteer computing systems
The Journal of Supercomputing
| Hi-index | 0.00 |
The BOINC toolkit is a very successful software infrastructure for the development of "volunteer computing" projects. For data distribution, the standard BOINC architecture provides uses a centralized server, or fixed set of pre-configured servers, each with a copy of a project's input files. This mirrored configuration has provided a way for BOINC projects to increase the size of their user-base and manage their data as projects grow. However, such a centrally administered approach can be limiting in terms of performance metrics, such as scalability and reliability, and also for management, since such servers have to be manually installed and configured for use. In this paper, we study an alternative approach, which is capable of providing a more lightweight and dynamic environment for distributing data, called Attic. Attic makes use of a secure decentralized approach and BitTorrent-like file swarming techniques to serve data and manage load. To compare our approach to the current centralized infrastructure, we show three experiments where we have integrated Attic with BOINC, each compares how Attic performs against BOINC as fragments (chunks) of the original data file are distributed and downloaded concurrently from multiple data centers. Our results show that careful choice of chunk size, by analyzing a server's uploading capabilities and file size, can result in large throughput gains for Attic when compared to the current BOINC data distribution paths.