In this article, an NAT-compatibility testbed is designed to automatically conduct the repeated experiments, collect the test results, and verify the direct connection rate (DCR) of any NAT traversal programs. Achieving a high DCR is important because using relays may unnecessarily increase the bandwidth cost, processing load of the relay servers, and the end-to-end packet delay. The NAT-compatibility testbed is constructed with 4 components: two peers, an automatic execution mechanism, NAT devices, and signaling/intermediate servers such as STUN, TURN, and SIP. It is also called the fully meshed testbed because the test result of all NAT combinations is a square. It measures the DCR of ICE, KeyStone, and PJNATH as 53.7%, 59.87%, and 50.93%, respectively. Experimental results show that asymmetric and unexpected direct connectivity check results occur in the real Internet. In order to enhance NAT traversal capability, the findings, like "port prediction" and "call-role sensitivity problem," are also described in the experimental results.
- NAT traversal
- direct connection