What is QoS?

QoS is called "Quality of Service" in English, and its Chinese name is "service quality ". QoS is a qualitative Convention on Information Transmission and sharing between users on the network and between users on the network. For example, the transmission delay allowed time, minimum transmission image distortion, and audio-video synchronization, it is a technology used to solve network latency and congestion problems. Currently, all routers support QoS.

Under normal circumstances, if the network is only used for a specific application system with no time limit, QoS is not required, such as web applications, or e-mail settings. However, it is necessary for key applications and multimedia applications. When the network is overloaded or congested, QoS ensures that important business volume is not delayed or discarded, while ensuring efficient network operation.
QoS has the following features:
1. Category
Classification means that a network with QoS can identify which application generates which data packets. Without classification, the network cannot determine the processing of special data packets. All applications will leave an identifier on the data packet that can be used to identify the source application. Classification is to check these identifiers to identify which application generates data packets. The following are four common classification methods.
(1) Some protocols are very "talkative". As long as they exist, they will lead to business latency. Therefore, identifying and prioritizing data packets based on the protocol can reduce latency. Applications can be identified by their ethertypes. For example, the appletalk protocol uses 0x809b and IPX uses 0x8137. Priority Processing Based on protocols is a powerful way to control or prevent the "talkative" protocols used by a few older devices.
(2) TCP and UDP port numbers many applications use some TCP or UDP ports for communication, for example, HTTP uses TCP port 80. By checking the port number of the IP packet, the SMART network can determine which application generates the packet. This method is also called layer-4 Switching because both TCP and UDP are located at the layer-4 of the OSI model.
(3) source IP addresses many applications are identified by their source IP addresses. Because the server is sometimes configured for a single application, such as an email server, the source IP address of the analysis data packet can identify the application that generates the data packet. This method is useful when identifying that a vswitch is not directly connected to an application server and that data streams from many different servers reach this vswitch.
(4) The physical port number is similar to the source IP address. The physical port number can indicate which server is sending data. This method depends on the ing between the physical ports of the vswitch and the application server. Although this is the simplest form of classification, it depends on the server directly connected to the vswitch.
2. Annotation
After identifying a data packet, you must mark it so that other network devices can easily identify the data. Because the classification may be very complex, it is best to perform it only once. After an application is identified, its data packets must be marked to ensure that the switches or routers on the network can process the application with priority. By adopting two industry standards for data labeling, namely IEEE 802.1p or differentiated service coding point (dscp), you can ensure that multi-vendor network devices can process the business with priority.
When selecting a vswitch or vro, make sure that it can identify two marking schemes. Although dscp can replace IEEE 802.1 p, which is dominant in LAN environments, it has some limitations compared with IEEE 802.1p. In a certain period of time, compatibility with IEEE 802.1 p devices will be very important. As a transitional mechanism, you should select a switch that can be switched from one solution to another.
3. priority settings
Once the network can distinguish between phone calls and online browsing, priority processing ensures that large downloads on the Internet are not interrupted. To ensure accurate priority processing, all traffic must be identified within the network backbone. Data priority processing on workstation terminals may be caused by human errors or malicious damages. Hackers can intentionally label common data as high-priority, and steal the bandwidth of important commercial applications, leading to invalid commercial applications. This situation is called a denial of service attack. By analyzing all traffic into the network, you can check security attacks and immediately stop them before they cause any harm.
In LAN switches, multiple service queues allow the existence of packet priorities. High-priority businesses can use vswitches without being affected by low-priority businesses to reduce latency incidents for time-sensitive businesses such as voice or video.
To provide priority, each port of the vswitch must have at least two queues. Although each port has more queues to provide more precise priority options, it is unlikely that more than four queues are required for each port in the LAN environment. When each packet arrives at the vswitch, it must be allocated to an appropriate queue based on its priority level, and then the vswitch forwards data packets from each queue. The switch uses its queuing mechanism to determine the queue to be served next. There are two queuing methods.
(1) strict priority queue (spq) is the simplest queuing method. It first serves the queue with the highest priority until the queue is empty, then, it will serve the next high-priority queue, and so on. The advantage of this method is that high-priority businesses are always processed before low-priority businesses. However, low-priority businesses may be completely blocked by high-priority businesses.
(2) Weighted Round Robin (WRR) is used to serve all business queues, and priority is assigned to higher-priority queues. In most cases, WRR processes high priorities first, but when there are many high-priority businesses, the lower-priority businesses are not completely blocked.

Article Referenced from: