The rotation of identifiers is a common security mechanism to protect telecommunication; one example is the frequency hopping in wireless communication, used against interception, radio jamming and interferences. In this thesis, we extend this rotation concept to the Internet. We use the large IPv6 address space to build pseudo-random sequences of IPv6 addresses, known only by senders and receivers. The sequences are used to periodically generate new identifiers, each of them being ephemeral. It provides a new solution to identify a flow of data, packets not following the sequence of addresses will be rejected. We called this technique “address spreading”. Since the attackers cannot guess the next addresses, it is no longer possible to inject packets. The real IPv6 addresses are obfuscated, protecting against targeted attacks and against identification of the computer sending a flow of data. We have not modified the routing part of IPv6 addresses, so the spreading can be easily deployed on the Internet. The “address spreading” needs a synchronization between devices, and it has to take care of latency in the network. Otherwise, the identification will reject the packets (false positive detection). We evaluate this risk with a theoretical estimation of packet loss and by running tests on the Internet. We propose a solution to provide a synchronization between devices. Since the address spreading cannot be deployed without cooperation of end networks, we propose to use ephemeral addresses. Such addresses have a lifetime limited to the communication lifetime between two devices. The ephemeral addresses are based on a cooperation between end devices, they add a tag to each flow of packets, and an intermediate device on the path of the communication, which obfuscates the real address of data flows. The tagging is based on the Flow Label field of IPv6 packets. We propose an evaluation of the current implementations on common operating systems. We fixed on the Linux Kernel behaviours not following the current standards, and bugs on the TCP stack for flow labels. We also provide new features like reading the incoming flow labels and reflecting the flow labels on a socket.