Search
Home Search Center IP Encyclopedia Online Courses Intelligent Model Selection

What Is AQM?

Application-network Quality Measurement (AQM) is a single-end measurement technology that dynamically adjusts the packet sending frequency based on the network quality and simulates application characteristics for proactive detection. It implements application-level refined network quality monitoring as well as real-time and visualized traffic management.

Why Do We Need AQM?

Commonly used audio and video conferences mainly utilize the Software as a Service (SaaS) deployment mode. Currently, no application-based high-precision measurement mode is available for uplink flows in WAN SaaS scenarios. If freezing or interruption occurs in audio and video conferences, it is difficult to demarcate and locate faults, resulting in poor user experiences for extended periods of time. The following discusses the advantages and disadvantages of Packet Conservation Algorithm for Internet 2.0 (iPCA 2.0) and Network Quality Analysis (NQA), which are major measurement technologies.

  • iPCA 2.0: is an in-band flow measurement technology deployed based on real services, which can accurately reflect the service quality. However, iPCA 2.0 is a dual-end measurement technology and cannot be effectively deployed on servers in SaaS scenarios. Instead, it can only detect the quality of a campus intranet.
  • NQA: is a basic measurement technology that supports diversified measurement types, and can be deployed on one end. However, NQA does not consider application characteristics. When the network Service Level Agreement (SLA) of an audio and video conference is actually measured, the transmission quality of the application cannot be accurately reflected due to the lack of parameter configuration and low measurement precision.

AQM can address these issues. It can restore traffic paths based on path discovery, and send packets by simulating characteristics of actual application flows such as the packet size and sending frequency. It can also demarcate and locate faults based on traffic paths and detection results, greatly improving measurement precision and shortening the fault locating time of key services from days to minutes. In addition, AQM can report key information such as the network topology, measurement indicators, and fault locations to the NMS and integrate the network digital map. This improves user experience and O&M efficiency, and helps build an intelligent O&M system.

AQM is applicable to midsize and large campus networks in various fields, such as manufacturing, education, and government, which brings the following benefits:

  • Network SLA monitoring before important conferences: Inspects and continuously monitors the quality of the network accessed by audio and video conference systems.
  • Fault demarcation during important conferences and routine O&M: Quickly demarcates any faults when a terminal accesses the conference system to prove network innocence or locate the network fault.
  • 24/7 network SLA monitoring for routine O&M: Monitors the access quality of routine office conferences on the enterprise headquarter and branch networks to quickly detect group faults and implement proactive O&M.

Application Scenarios of AQM

The following figure shows a typical application scenario of AQM, where AQM covers the wired campus network, WAN, and the last hop between edge devices (Cloud Edge) and the server. The measurement point is usually deployed at the core or aggregation layer to detect the quality of the network to servers of target applications.

Typical application scenario of AQM
Typical application scenario of AQM

How Does AQM Work?

The AQM process consists of four phases: path discovery, slow detection, fast detection, and hop-by-hop fault locating.

Path Discovery

In the path discovery phase, path detection is implemented based on UDP packets. The destination address to be detected is the server address, which can be manually specified or automatically obtained. For details, see Automatic Detection Address Identification. The path discovery process consists of campus network path discovery and WAN path discovery.

Table 1-1 Path discovery process

Detection Type

Detection Scope

Detection Process

Detection Effect

Campus network path discovery

A total of 16 paths from aggregation devices, core devices, online behavior management devices, firewalls, to Internet Service Provider (ISP) routers, assuming that all of these devices are active-active and fully meshed

UDP packets are constructed for detection. For a UDP flow, the TTL value starts from 1 and increases by 1 hop by hop. If the IP address of a hop matches that of an edge device on the campus network, the UDP flow detection ends. When timeouts occur at 10 consecutive hops or the TTL value reaches the upper limit, the UDP flow detection ends.

When the number of discovered paths reaches 16, or topology convergence occurs (that is, the upper limit of detection flows inferred hop by hop based on the topology is reached), the path discovery ends.

Displays a complete topology on the campus network, including 16 paths and UDP port features required for covering these 16 paths.

WAN path discovery

Three paths selected from ISP routers to Cloud Edges

UDP packets are constructed for detection. For a UDP flow, the port number of the campus network path is reused. The TTL value starts from that of a campus edge device and increases by 1 hop by hop. One detection packet is sent at each hop. When timeouts occur at 10 consecutive hops or the TTL value reaches the upper limit, the UDP flow detection ends.

A maximum of 16 flows are supported for detection. Multiple paths may be detected, from which three paths are selected. Then the path discovery ends.

Covers three paths from the first hop of the enterprise egress to the last hop to the cloud.

Path discovery is performed at a fixed interval. If the topology does not change, subsequent detection behaviors remain unchanged. If the topology changes, subsequent detection behaviors are determined based on the new topology. If path discovery fails, the current detection is stopped.

The intelligent multi-path discovery algorithm can optimize equal-cost multi-path routing (ECMP) measurement precision and periodically update the topology, improving the overall accuracy. For details, see Intelligent Multi-Path Discovery Algorithm.

Slow Detection

After the path discovery phase is complete, slow detection is triggered. In this phase, detection is implemented based on ICMP packets. A detection session contains a maximum of four detection flows. The destination addresses that can be detected include the server address and a maximum of three Cloud Edge addresses. The packet size of each detection flow is 972 bytes (payload of the IPv4 type). Packets are evenly sent at a low frequency and at a fixed interval.

The detection result of the server address is preferentially used as the slow detection result. If the server address is unreachable for a long time, the detection results of Cloud Edge addresses are compared based on the packet loss rate first and then the delay for selecting the worst result. That is, if multiple detection flows have measurement indicators reaching thresholds, the detection result with the higher packet loss rate is preferentially selected. When the detection result reaches the threshold, fast detection is triggered.

Fast Detection

In the fast detection phase, a detection session contains only one detection flow. The destination address to be detected is the address for which a slow detection threshold alarm is triggered. The packet size, packet sending frequency (packets are sent unevenly in geometric distribution mode), and DSCP value of a detection flow are set based on characteristics of simulated applications, covering camera, screen sharing, and audio applications. For details about application characteristics, see Application Characteristic Extraction. Fast detection supports low-precision and high-precision measurement modes. By default, the low-precision measurement mode is used. You can manually switch to the high-precision measurement mode. In high-precision measurement mode, as the statistical period is prolonged, the CPU usage increases sharply and more bandwidth is occupied.

The rule for selecting the worst detection result for fast detection is the same as that for slow detection. When the detection result reaches the threshold, hop-by-hop fault locating is triggered. In addition, fast detection suppression is enabled for the detection flow for a period of time.

Hop-by-Hop Fault Locating

In the hop-by-hop fault locating phase, network faults can be demarcated by segment and faults on the campus network can be located hop by hop based on the path discovery result and fast detection model.

Table 1-2 Segment-based fault demarcation + hop-by-hop fault locating process

Detection Type

Demarcation/Locating Result

Segment-based fault demarcation

Whether a fault occurs on the wired campus network, WAN, or the last hop of the server:
  • Detect the address of an ISP router and check whether the packet loss rate or delay meets the following thresholds. If so, the fault has occurred within the campus network.
    • Packet loss rate: 50% of the packet loss rate threshold of fast detection for triggering hop-by-hop fault locating is used as a standard.
    • Delay: It is the same as the delay threshold of fast detection.
  • Detect the address of a Cloud Edge. If hop-by-hop fault locating is triggered and no fault occurs within the campus network, the fault has occurred on the WAN.
  • Detect the server address. If hop-by-hop fault locating is triggered and no fault occurs on the campus network or WAN, the fault has occurred on the last hop of the server.

Hop-by-hop fault locating within the campus network

If a fault occurs within the campus network, hop-by-hop fault locating is performed along the paths discovered in the path discovery phase to check whether each hop address of each path is reachable. The detection model for hop-by-hop fault locating is the same as the fast detection model. Considering the detection performance, if the statistical period in this phase is shorter than that in the fast detection phase, the overall measurement precision in this phase is lower.

What Are the Key Technologies of AQM?

Automatic Detection Address Identification

Automatic detection address identification includes server address obtaining and server address deletion.

  • Server address obtaining: A server address can be statically configured or dynamically learned. The implementation process is as follows:
    • Static configuration: A user obtains the characteristics of service flows in advance and manually specifies the server address when delivering an AQM instance.
    • Dynamic learning: A user does not specify the server address when delivering an AQM instance. In this case, AQM can use the application identification module to accurately identify different applications and automatically obtain the server address.
  • Server address deletion: After an application goes offline, the application identification module notifies the user and deletes the server address.

Intelligent Multi-Path Discovery Algorithm

In ECMP scenarios, the common path discovery algorithm does not consider the per-flow load balancing policy of devices. As a result, topology path information may be incorrectly or incompletely detected. AQM uses an industry-unique intelligent multi-path discovery algorithm, which constructs multiple detection flows with different 5-tuple information (source address, destination address, source port number, destination port number, and protocol type) to discover paths. The algorithm has the following characteristics:
  • The load balancing hash field of the same detection flow remains unchanged to ensure that packets of the same detection flow are transmitted along the same forwarding path, preventing topology path information detection errors.
  • Before the number of paths reaches the upper limit or topology convergence occurs, all paths for load balancing are discovered using detection flows, preventing incomplete detection of topology path information.
Comparison between the traditional path discovery algorithm and intelligent multi-path discovery algorithm
Comparison between the traditional path discovery algorithm and intelligent multi-path discovery algorithm

Application Characteristic Extraction

For received packets, packet loss occurs randomly in the queue. When detection packets and application packets enter the queue at the same time, and a small number of detection packets are transmitted, it is more difficult to discard detection packets. As a result, the measured packet loss rate decreases. Similarly, when a large number of detection packets are transmitted, the measured packet loss rate increases. The measured packet loss rate can be closest to the actual packet loss rate of applications only when the number of detection packets is the same as that of application packets and sufficient data statistics are collected.

As such, in the fast detection phase of AQM, you can adjust the number of concurrent packets sent instantaneously according to different application and network characteristics. You can also simulate packet sending based on application characteristics to approximate to the transmission process of real application packets, thereby obtaining more accurate network transmission quality.

Simulating the characteristics of real application packets
Simulating the characteristics of real application packets

In addition, you need to construct an application characteristic table of well-known applications, simulate network congestions through fault injection, and extract key application characteristics. The preset top applications include the following: WeLink_Meeting, TencentMeeting, DingTalk_VoIP, XiaoYuYiLian, Zoom_Video, MicrosoftTeams_VOIP, WebEx_VoIP, and Lark_VoIP.

About This Topic
  • Author: Chen Jingyi
  • Updated on: 2025-02-05
  • Views: 1043
  • Average rating:
Share link to