A Hybrid Load Balancing Method in Data Center Network

Abstract

The invention discloses a hybrid load balancing method in a data center network, in which a central controller is deployed to obtain network global information, terminal hosts are deployed to perform fast routing of small and medium flows, and at the same time, the central controller is given the decision-making power of centralized load balancing And the decision-making power of distributed load balancing of terminal hosts; the central controller uses global information to arrange paths for predefined large flows, and calculates the globally optimal network queuing delay for predefined small and medium flows, and the terminal hosts Distributed route selection is performed on the small and medium streams, and rerouting is performed under the limitation of the global optimal network delay. Let the central controller and the terminal host cooperate with each other to ensure the high throughput of large flows and the low-latency requests of small and medium flows at the same time in the network situation where different traffic requirements coexist. Compared with a simple centralized or distributed scheme, the method of the invention can take into account all traffic in the network, and meet the requirements of different traffic, and is flexible, convenient, and easy to configure.

Description

technical field [0001] The invention relates to the field of computer networks, in particular to a hybrid load balancing method in a data center network. Background technique [0002] The Internet provides humans with an extremely convenient platform for information communication and resource sharing, profoundly affects people's production, life and learning methods, and has also become an important infrastructure supporting production development, economic development and technological innovation in modern society. With the rapid development of the Internet, various application services have emerged as the times require. From real-time applications such as web search and recommendation systems to high-throughput services such as data backup and virtual machine migration, all of these impose strict requirements on network data transmission. As a dedicated high-speed network, the data center network provides a high-performance platform for diverse services. Various applicat...

AI Technical Summary

Problems solved by technology

However, the disadvantages of the centralized scheme are also obvious, mainly as follows: (1) The global information collection, path calculation and rule distribution of the central controller have a time cost. Controller computed path is unreasonable

(2) The central controller collects the information of the whole network and needs to exchange information with the switch. If the interaction is too frequent, a large amount of additional traffic overhead will be introduced

(3) It is extremely dependent on the central controller and has poor scalability. Once the controller fails, the entire system will not work

The distributed load balancing scheme for obtaining global information usually requires the cooperation of switches for global information feedback, which will either introduce a large amount of additional traffic overhead (such as DADR, combined with openflow, to collect information on each terminal), or need to modify the switch hardware to achieve Information collection requirements (such as CONGA, HULU, etc., use the switch to mark the CE mark on the header), although this type of solution can achieve better load balancing effects, and maintain the characteristics of fast and scalable distributed solutions, but their Feedback systems are often very complex, and current commercial switches cannot support these functions

[0008] In general, both the centralized load balancing scheme and the distributed load balancing scheme have obvious defects, they either ignore the impact of small and medium flows on the network, or ignore the importance of global decision-making

Images