Chapter 8
Conclusions and future work
This work focused on the SIP-based VoIP integration with the peer-to-peer paradigm, in particular when SIP User Agents are behind a NAT device.
SIP User Location Service needs to retrieve the contact address associ-ated to a unique URI (the Address Of Record). This kind of precise search reveals the inadequacy of unstructured peer-to-peer overlays to implement a distributed User Location Service, but rather makes clear the need to employ a Distributed Hash Table (DHT) algorithm.
Major DHT algorithms have been briefly described, analyzing their dis-tinctive features. It is important to, note, however, that literature versions of those algorithms are actually different from their actual and largely de-ployed implementation (when existent), because they have been undergoing an “on-the-road” tuning. Anyway, a P2P SIP UA should be independent on the actual DHT protocol, and allow easy interchange of it.
NAT operations have been examined and NAT traversal techniques have been analyzed. In order to provide a universally working NAT traversal al-gorithm, those techniques have been in particular considered, which do not make assumptions on the NAT device and require only augmented intelli-gence in end-devices.
The ALEX SIP extension to improve direct connectivity in SIP have been examined and extended to work also when NAT devices are along the network path. ALEX automatically activates the considered NAT traversal
CHAPTER 8. CONCLUSIONS AND FUTURE WORK techniques.
Most NAT Traversal algorithms require public servers that relay traffic on behalf of masqueraded nodes and act as a rendez-vous point. In a true peer-to-peer environment, however, no centralized public servers are available. Therefore, a Distributed Relay Service protocol has been design in order to allow peers with a public address to offer such service for nodes behind NAT. Care have been devoted to ensure reachability of masqueraded peers, redun-dancy of service and distribution of load across public peers. Those peers are advertised through a gossip-based unstructured overlay, whose information is piggy-backed in regular SIP messages. Moreover a feed-back mechanism from relay peers to masqueraded ones has been designed, in order to allow load level and graceful shutdown notification.
It is important to underline that our protocol is RFC-compliant. Sub-scription and URI semantics have not been modified. Just one new event package has been defined. Moreover, peers that do not support our mecha-nism will automatically reply with a standard error message. This protocol has been carefully designed to ensure backwards compatibility and incremen-tal implementation of the functions into the nodes.
In the end, statistics on NAT diffusion and peer lifetime of world-wide peer-to-peer networks have been used to analyze the efficiency of our algo-rithm. The unstructured overlay was tuned by simulating it under different parameter values, and proved to achieve good distribution of knowledge of the public peers which provide relay service.
The only critical value is the percentage of hosts behind NAT, which has been measured as high as 74%. This is a limiting factor not only for our algorithm, but for any fully decentralized overlay which aims at providing efficient NAT traversal for its peers. This number is not likely to lower; its negative effects, however, is going be mitigated by the lowering percentage of Symmetric NATs in favor of Full Cone or Restricted NATs. Employment of IPv6 can be advertised as a solution to NAT problems; however, this is happening at a too slow rate.
In the mean time, NAT-cooperating techniques (UPnP, NAT-PMP, . . . ) can be explored in order to allow peer to gain a public IP address whenever
CHAPTER 8. CONCLUSIONS AND FUTURE WORK possible without the help of other nodes.
8.1
Future work
Although it has been validated with statistics and simulation, the solution proposed in this work needs to be implemented and tested on an as much as possible large scale.
In second place, security must be addressed. As there is no centralized authority that can authenticate users, UA should obtain a certificate from a well-known Certification Authority and use their private key to sign SIP messages, according to the S/MIME protocol, for example. In this way other peers can autonomously verify the integrity and authenticity of messages.
In third place, the P2P SIP approach can be seen from a broader point of view than this work did. In fact, P2P SIP can be detached form the tra-ditional VoIP application, and assume the role of a generic framework for decentralized user location and end-to-end connection establishment. Those connections can be used for any purpose: file transfer, video streaming, shared whiteboard, remote desktop, etc., and, obviously audio/video calls and instant messaging.
