What Is Intra-AS Bandwidth Pooling?

Why Do We Need Intra-AS Bandwidth Pooling?

Network traffic has become more and more diverse and dynamic with the emergence of many new services, such as cloud computing, big data, and the Internet of Things (IoT). This increases the need for flexible and efficient resource management. While existing technologies, including Equal-Cost Multi-Path (ECMP), Unequal Cost Multi Path (UCMP), link-group, and ECMP Fast Reroute (FRR), can adjust network traffic paths, each has its own limitations.

Although there are various traffic adjustment technologies, they each struggle with complex scenarios. For example, if multiple parallel load-balancing links are used between two devices and some parallel links of the primary link fail or traffic surges occur, the devices cannot automatically detect the faults or congestion. This leads to traffic congestion and packet loss. When using these technologies, the system faces the following challenges:

• The system lacks automatic fault detection capabilities and identifies issues only through passive monitoring.
• Locating faults on a large network can take several hours or more.
• After a fault is located, the configuration must be manually modified to re-allocate traffic paths. This is slow, prone to errors, and requires skilled O&M personnel. A more efficient and user-friendly solution is needed.

Intra-AS bandwidth pooling is introduced to address these issues. It detects faults in real time, quickly and automatically adjusts traffic paths, and relieves congestion within seconds.

What Is Intra-AS Bandwidth Pooling?

Intra-AS bandwidth pooling groups available bandwidth within the same IS-IS AS for unified management, enables devices to proactively learn network topology and link traffic information through IS-IS, and adaptively optimizes traffic paths based on actual traffic amounts and available bandwidth. This enables unequal-cost load balancing and relieves congestion within seconds.

Intra-AS bandwidth pooling offers the following functions:

• Collects and advertises real-time bandwidth usage and traffic statistics on interfaces at intervals (as short as 10s) through IS-IS.
• Analyzes traffic on all intra-AS links, quickly identifies congested paths, and automatically calculates available backup paths and optimization ratios based on the topology and real-time traffic information using AI algorithms. This provides precise guidance for quickly easing congestion.
• Implements unequal-cost load balancing based on the optimization result to steer traffic intelligently and ease link congestion.
• Automatically senses a link recovery and switches traffic back to the original path.

How Does Intra-AS Bandwidth Pooling Work?

Concepts related to intra-AS bandwidth pooling:

• Congestion threshold: If the bandwidth usage reaches or exceeds this threshold, the network is considered congested, and path optimization starts.
• Maximum bandwidth usage of an optimization path: After path optimization is performed, the bandwidth usage of an adaptive optimization path should not exceed this value.
• Bandwidth usage for exiting path optimization: The system calculates the primary link's bandwidth usage under the assumption that this link carries all traffic. If the calculated usage stays at or below the bandwidth usage for exiting path optimization within the delay for exiting optimization, the system terminates path optimization and triggers traffic switchback to the primary link.
• Delay for exiting path optimization: The system waits to switch traffic back if the primary link's bandwidth usage stays at or below the bandwidth usage for exiting path optimization until the delay ends.

The following figure shows a topology where IS-IS neighbor relationships are established between devices, with two parallel load-balancing links deployed between Device A and Device C. IS-IS bandwidth advertisement is configured on each device so that neighboring devices can obtain the bandwidth usage of each link. Basic SRv6 functions and advertisement of parallel End.X SIDs are configured on each device, ensuring that parallel links can be generated between devices.

This network has three paths from Device A to Device C: Path 1 (Device A -> Device C), Path 2 (Device A -> Device B -> Device C), and Path 3 (Device A -> Device D -> Device C). Path 1 is the primary path because it is the shortest. Paths 2 and 3 are backups, known as adaptive optimization paths.

Path 1 has two parallel links, each with a bandwidth of 1 Tbit/s, giving a total bandwidth of 2 Tbit/s. The traffic volume is 1.1 Tbit/s. If one of these links fails, the available bandwidth drops to 1 Tbit/s. Although the traffic forwarding path remains unchanged (the other link is still available), the traffic exceeds the available bandwidth. This leads to traffic congestion and packet loss.

Typical networking of intra-AS bandwidth pooling

With intra-AS bandwidth pooling enabled, the system uses IS-IS to learn traffic and available bandwidth information in real time, quickly detect congestion, and optimize paths. The principles are as follows:

• Optimization rules:
  • When the system detects that the bandwidth usage of Path 1 exceeds the congestion threshold, it adds Path 2 (Device A -> Device B -> Device C) and Path 3 (Device A -> Device D -> Device C) as adaptive optimization paths to the primary path list. It then steers some traffic to Path 2 and Path 3 according to the bandwidth they have. This achieves unequal load balancing among the original and new paths, thereby avoiding congestion on the original primary path.
  • The system uses AI algorithms to calculate the load balancing weight. It considers the total traffic on the primary path, available bandwidth of each adaptive optimization path, and maximum bandwidth usage of each adaptive optimization path. This ensures that the adaptive optimization paths are not congested after some traffic is steered to them while avoiding congestion on the primary path.

• Traffic switchback rules:
  • After path optimization is triggered, IS-IS keeps monitoring Path 1's bandwidth usage in real time. If the link fault is rectified or the traffic volume drops and IS-IS determines that switching all traffic back to Path 1 would not congest Path 1 (specifically, Path 1's bandwidth usage would stay below the bandwidth usage for exiting path optimization), IS-IS starts the timer for exiting path optimization. If Path 1's bandwidth usage stays below the bandwidth usage for exiting path optimization during the timer period, traffic is switched back to Path 1 when the timer expires.