Virtual Network Functions as a Service (VNFaaS) is currently under attentive study by telecommunications and cloud stakeholders as a promising business and technical direction consisting of providing network functions as a service on a cloud (NFV Infrastructure), instead of delivering standalone network appliances, in order to provide higher scalability and reduce maintenance costs. However, the functioning of such NFVI hosting the VNFs is fundamental for all the services and applications running on top of it, forcing to guarantee a high availability level. Indeed the availability of an VNFaaS relies on the failure rate of its single components, namely the servers, the virtualization software, and the communication network. The proper assignment of the virtual machines implementing network functions to NFVI servers and their protection is essential to guarantee high availability. We model the High Availability Virtual Network Function Placement (HA-VNFP) as the problem of finding the best assignment of virtual machines to servers guaranteeing protection by replication. We propose a probabilistic approach to measure the real availability of a system and design both efficient and effective algorithms that can be used by stakeholders for both online and offline planning.