Does an AR router support link aggregation or load balancing when it accesses the Internet through multiple upstream interfaces?
If the multiple upstream interfaces are physical interfaces of the same type (such as GE interface or Eth-Trunks), the bandwidth of these interfaces can be aggregated, which can be deemed equivalent to an Eth-Trunk. The bandwidth of the aggregated interface equals to the sum of the bandwidths of the physical links. For example, if two Eth-Trunks are aggregated, the total bandwidth of the aggregated interface is 200 Mbit/s (100 Mbit/s + 100 Mbit/s).
However, if the multiple upstream interfaces are service interfaces, the total bandwidth after aggregation cannot reach the sum of the bandwidths of the interfaces. For example, if two broadband interfaces (4 Mbit/s and 2 Mbit/s respectively) provided by different carriers are aggregated, the rate of the aggregated bandwidth cannot reach 6 Mbit/s. In addition, if the two upstream interfaces work in ECMP load balancing mode, the rate of the aggregated interface further decreases and may even be lower than the rate of a single upstream interface. The reason is that the interfaces provided by two carriers belong to different networks, whose packet transmission latency and jitter are different. In this case, the response packets corresponding to TCP link packets (which are used by most services) are disordered, resulting in packet reassembly, packet retransmission, or even the disconnection and re-establishment of TCP links. As a result, the network access rate is low or users cannot open web pages. You can perform the following measures to resolve the problem:
Use specific routes to distinguish services. Do not use ECMP or use only one upstream interface.
Use traffic policies to redirect the next hop to distinguish services, and do not configure load balancing for services.