Modern large-scale networked services, such asvideo streaming, are typically deployed at multiple locationsin the network to provide redundancy and load balancing.Different techniques are used to provide performance monitoringinformation so that client nodes can select the best serviceinstance. One of them is collaborative sensing, where clientsshare measurement results on the observed service performanceto build a common ground of knowledge with low overhead.Clients can then use this common ground to select the mostsuitable service provider. However, collaborative algorithms aresusceptible to false measurements sent by malfunctioning ormalicious nodes, which decreases the accuracy of the performancesensing process. We propose Sense-Share, a simple light-weightand resilient collaborative sensing framework based on the simi-larity of the client nodes’ perception of service performance. Ourexperimental evaluation in different topologies shows that serviceperformance sensing using Sense-Share achieves, on average,94% similarity to non-collaborative brute force performancesensing, tolerating faulty nodes. Furthermore, our approacheffectively distributes the service monitoring requests over theservice nodes and exploits direct inter-node communication toshare measurements, resulting in reduced monitoring overhead

Peer Reviewed