A fault tolerant, peer-to-peer based scheduler for home grids

This thesis presents a fault-tolerant, Peer-to-Peer (P2P) based grid scheduling system for highly dynamic and highly heterogeneous environments, such as home networks, where we can find a variety of devices (laptops, PCs, game consoles, etc.) and networks.

The number of devices found in a house that are capable of processing data has been increasing in the last few years. However, being able to process data does not mean that these devices are powerful, and, in a home environment, there will be a demand for some applications that need significant computing resources, beyond the capabilities of a single domestic device, such as a set top box (examples of such applications are TV recommender systems, image processing and photo indexing systems). A computational grid is a possible solution for this problem, but the constrained environment in the home makes it difficult to use conventional grid scheduling technologies, which demand a powerful infrastructure.

Our solution is based on the distribution of the matchmaking task among providers, leaving the final allocation decision to a central scheduler that can be running on a limited device without a big loss in performance.

We evaluate our solution by simulating different scenarios and configurations against the Opportunistic Load Balance (OLB) scheduling heuristic, which we found to be the best option for home grids from the existing solutions that we analysed. The results have shown that our solution performs similar or better to OLB. Furthermore, our solution also provides fault tolerance, which is not achieved with OLB, and we have formally verified the behaviour our solution against two cases of network partition failure.