Method, device and network system for realizing load balancing

Abstract

Embodiments of the present application provide a method, device and network system for implementing load balancing. The controller obtains the port attribute of each port of the first switch in the mesh network; the port attribute user side port and the network side port; according to the port attribute of each port of the first switch, determine the first switch Each port of the corresponding virtual route forwarding instance; wherein, all user-side ports of the first switch correspond to the first virtual route forwarding instance; according to the topology of the mesh network and the virtual route of each port of the first switch A forwarding instance, generating a forwarding table of the first switch, each forwarding entry in the forwarding table includes a virtual routing forwarding instance, when the virtual routing forwarding instance is the first virtual routing forwarding instance, the The forwarding table entry further includes a destination network segment and at least two forwarding paths corresponding to the destination network segment; and sending the forwarding table to the first switch.

Description

technical field [0001] The present application relates to network communication technologies, and in particular to a method, device and network system for realizing load balancing. Background technique [0002] The network data traffic model in the era of cloud computing has undergone great changes. 16% of the data traffic is on the carrier network, 14% is on the enterprise network, and the remaining 70% of the data traffic will all flow inside the data center network. The exponential growth of data traffic and bandwidth in data center networks has far exceeded people's imagination of traditional networks. Large bandwidth and high scalability have become the primary demands of data center network customers. The bottleneck of data center network construction is becoming more and more obvious. [0003] In order to overcome the above problems, a scalable data center network, called a mesh network, is provided in the prior art. A mesh network refers to a fully connected netwo...

AI Technical Summary

Problems solved by technology

This causes the shortest path between the first switch and the second switch to be overloaded, while other paths between the first switch and the second switch are underutilized

If it is necessary to achieve load balancing in multiple unequal-cost paths, it is necessary to release the restriction of the shortest path first principle. In this case, since all switches in the mesh network are fully connected, and the existing forwarding table Each table entry in the mesh network only includes the destination network segment, next hop and port. The switch that receives the message in the mesh network randomly selects a forwarding path for the message according to the forwarding table, and the message may be passed The source switch (that is, the switch connected to the server that sent the message) continues to forward until the time to live (English: time to live, TTL) of the message is 0, and the message is discarded

It can be seen that only letting go of the restriction of the shortest path first principle cannot achieve load balancing in the mesh network

Images