Virtual machine distribution method in data center with minimized communication delay

Abstract

The invention discloses a virtual machine distribution method in a data center with minimized communication delay. The method comprises the following steps of S1, judging whether virtual machine colonies which can communication with one another and of which the scale is more than that of data nodes exist in a certain threshold of the communication delay; and S2, if yes, finding a complete subgraph composed of virtual machines under such threshold by using an efficient backtracking method or an improved greedy algorithm, and then adopting a Hopcroft-Karp algorithm to distribute the virtual machines in the complete subgraph to the data nodes; and if no, repeating the step S1. According to the virtual machine distribution method of the invention, solution space can be effectively reduced, and the virtual machine distribution time can also be reduced.

Description

technical field [0001] The invention relates to the field of virtual machine allocation, in particular to a virtual machine allocation method in a data center that minimizes communication delay. Background technique [0002] With the rapid development of data centers and the increasing importance of the Mapreduce / Hadoop framework, more and more big data are being processed in cloud systems. In modern virtualization-based data centers, virtual machine allocation is the primary consideration for efficient resource scheduling in the cloud. In cloud systems, big data is divided into multiple pieces of data and stored on data nodes in the data center to be processed by virtual machines. At this time, there is not only a communication delay when the virtual machine processes data, but also a communication delay between virtual machines when summarizing the calculation results. Different virtual machine allocation strategies will result in different maximum communication delays. ...

AI Technical Summary

Problems solved by technology

[0004] There are currently two methods to solve this problem. The first proposed heuristic algorithm is to use the greedy algorithm to find the candidate group of virtual machines for the data nodes and combine the Hungarian algorithm to allocate virtual machines for the data nodes. There are candidate The problem that the group size may be small or the matching efficiency is low

The second is to convert it into an approximate problem by changing the constraints of the original problem, and then relax it into a linear programming problem and use a linear programming solver to solve it. The solution space is still the solution space of the original problem, and a larger solution space will increase the search time. The time of better solution, the efficiency is not high

Images