The demise of Transport Layer

The idea of recent reading. The transport layer might exist even no longer than ipv4,

Since I started to learn the network and transport layer, there is always a question that why transport layer contains so many functions, including port, reliable transmission and congestion control. These three functions, to some extent, has no need to work on a single layer.

As Ipv6 replaces ipv4 gradually, in my opinion, the time for transport layer has also gone. TCP is being discarded in more and more scenarios. On the one hand, QUIC/HTTP3 uses UDP directly to bypass the middlebox and firewall, at the same time to implement several functions including congestion control and encryption in the user space. On the other hand, in data center, TCP is accused of CPU overhead and high latency from network stack. RDMA, especially RoCE (even not iWarp)_ gradually dominant the traffic. Interestingly, RoCEv2 adopts UDP/IP to transmit on Ethernet.

In the future, nobody would continue using TCP, a classical transport layer protocol living more than 40 years. Instead, people would use UDP to identify the port, and put other functions to higher layer.

But UDP is too simple. How could it survive when TCP has died? Why don’t we simply put the src port and dst port into ipv6 extended header?

Then transport layer dies.

But transport layer will not be dead indeed. The spirit and talent idea created before would live on the other layers.

The transport layer is decomposed.