A brief discussion on "lossless network": ECN and PFC technology

Basic concepts of lossless network

First of all, the most core feature of data center is lossless network. Traditional data center network presents three separate networks, that is, high-performance computing network, storage network, and general computing network use different protocols and interfaces.

As the Ethernet rate exceeds 800G and the Ethernet-based RDMA protocol (RoCEv2) is applied, the integration of three networks has become the development trend of data centers, and high throughput, low latency, and zero packet loss have also become the requirements of current data centers. These three requirements are called lossless network characteristics.

Disadvantages of high-speed forwarding in traditional networks

Based on the lossless network characteristics of the data center, some functional requirements are also put forward for the equipment. When data is forwarded on the Ethernet, when there is congestion and packet loss caused by burst traffic, it will affect the network transmission delay. The typical topology is shown in the following figure:

In order to reduce latency and solve the packet loss problem, Ethernet introduced flow control technology. However, traditional flow control technology has two obvious defects:

• First, it is based on ports and cannot distinguish traffic of different priority services.
• The second is the transmission characteristics, which affect the throughput of the entire network system.

Working defects of traditional fluid control technology:

To solve these two problems, switches introduce two functions to meet lossless requirements: PFC and ECN.

How lossless networks work

PFC (Priority-based Flow Control) adds the concept of priority to the traditional flow control technology. It can distinguish the traffic of services with different priorities and meet the requirements of ensuring high-priority service traffic. It is a point-to-point protocol. It truly realizes the sharing of links by multiple traffic. However, the problem of reduced system throughput caused by flow control transmission cannot be avoided.

ECN (Explicit Congestion Notification) was created to solve this problem. It actively reduces traffic by notifying the sender before triggering flow control, thus avoiding affecting the throughput of the entire network. It is an end-to-end protocol.

The combined effect of these two functions ensures lossless transmission of data in Ethernet.