When a network switch interface receives more traffic than it can handle, the network switch will choose to either buffer it, or the network switch will drop it.
The cache of the network switch is usually caused by the different speeds of the network interfaces, the sudden burst of traffic on the network switch or the many-to-one traffic transmission.
The most common problem that causes buffering in network switches is sudden changes in many-to-one traffic. For example, an application is built on multiple server cluster nodes. If one of the nodes requests data from the network switches of all other nodes at the same time, all replies should arrive at the network switches at the same time. When this happens, traffic floods from all the network switches to the port of the requester's network switch. If the network switch does not have enough egress buffers, the network switch may drop some traffic, or the network switch increases application latency. Adequate network switch buffers can prevent packet loss or network delays caused by low-level protocols.
Most modern data center switching platforms solve this problem by means of a shared switching cache of network switches. Network switches have a pool of buffer space allocated to specific ports. Network switch shared switch cache varies widely between vendors and platforms.
Some network switch vendors sell network switches designed for specific environments. For example, some network switches have large buffer processing, which is suitable for Hadoop environments in many-to-one transmission scenarios. Network switches In environments where traffic can be distributed, network switches do not require buffers at the switch level.
The buffer of the network switch is very important, but there is no correct answer to how much network switch space we need. Huge network switch buffers mean that the network doesn't drop any traffic, but it also means increased network switch latency—data that needs to wait before being forwarded by the network switch. Some network administrators prefer smaller network switch buffers, letting the application or protocol handle some of the traffic. The correct answer is to understand the traffic patterns of your application's network switches and choose a network switch that fits those needs.
