CAPS: A cloud-assisted approach to handle spikes in peer-to-peer web search

Abstract

In search engines, popular news events cause huge spikes in the queries related to the events. In this work, we consider the stability issues caused by these spikes in a peer-to-peer web search engine formed using voluntary resources shared by peers. The requirement of providing top-ranking results from a dynamic index distinguishes web search from other classes of peer-to-peer search like object lookup and file search. This makes the traditional methods of load reduction based on caching and replication, proposed for peer-to-peer object lookup and file search, insufficient for containing interest spikes in peer-to-peer web search. We propose transient use of public cloud to maintain the stability of peer-to-peer web search engines during interest spikes. To the best of our knowledge, this is the first work proposing transient use of public cloud to handle spikes in peer-to-peer search. In the proposed architecture, CAPS, the responsibility of handling spiking queries is dynamically offloaded to public clouds during the spike period. The peer bandwidth to be used for transfer of relevant index from peers to cloud is decided considering the impact on other applications in the peer as well as the requirements of the search application. We show that transient use of public cloud can be performed without major adverse impact on the desirable properties of peer-to-peer search like privacy and decentralized control. Experimental evaluation under realistic settings show that cloud-assistance can be used effectively to handle spikes.