Sensor-cloud was introduced to create a cloud environment where a geographically distributed pool of sensors (physical or logical) and cloud resources can be shared between numerous applications. In addition, sensor-cloud hides the complexity and heterogeneity of underlying physical/logical sensors from applications. Based on the applications' requirements, the sensor-cloud creates virtual sensors that provide an abstract layer for applications to interact with real sensors through a model. Each virtual sensor abstracts one given real sensor. In this work, we focus mainly on the performance aspect of sensor-cloud infrastructure. In fact, since same data can be requested by applications, if the sensor-cloud allocates several real sensors and creates multiple virtual sensors to retrieve it, many resources will be wasted. To overcome this problem, we propose to share real and virtual sensors between different applications based on their requirements to minimize the network and system resources utilization. The proposed mechanism permits indeed to minimize the number of instantiated virtual sensors in the sensor-cloud, as well the traffic to/from the physical/logical sensors while achieving the QoS objectives of applications. The performed simulations, show that by adopting this strategy, a further optimization of resources within the sensor-cloud infrastructure is possible without compromising the requested QoS.