Research On Techniques Of Network Performance Measurement In The Data Center Environments

With the rapid development of computer technology and the increasing demand forcomputer network, data centers, as the basis of a centralized data management and all ITservices, the network performance is directly related to the implementation of the entirenetwork service and the quality of the implementation. Timely and accuratemeasurement of key performance parameters of the network in the data centerenvironment is the premise of effective maintenance and management, optimaladjustment of flow strategy and the enhancement of the quality of the data centerrunning. There are many parameters related to data center network performance.Among them the measurements of availability (connectivity), network bandwidth anddelay are the key technology in the current data center network resource managementfield.Part of the data center network environment is relatively complex, which is mainlyreflected in the fact that most internal networks consist of a variety of networkconvergence. The system operating environment studied in this paper is a networkstructure that interconnects data center network (fusion IB network and Ethernet) andthe Internet. Compared to single network structure, the measurement of this networksystem needs to measure not only the performance of each subnet in the data centernetwork, but also the performance of any node within the data center server and Internet,and thus cause difficulty in measuring.This paper mainly studies the following aspects: Firstly, the background andsignificance of data center network performance measurement. Secondly, networkperformance measurement techniques, including the parameters, methods, tools formeasurement and related algorithms, with a focus on the difficulties of measuring thenodes within the data center internal network and any node between the data centerserver and Internet; as well as a summarization of the procedures of measuring datacenter network performance. Thirdly, given the fact that node collaboration can not beobtained when measuring the performance of any nod between the data center internalserver and Internet, Pathneck is chosen as the network measurement tool. The papergives detailed analysis of the functionality of Pathneck and the measurement principlethis tool; its operability is proven by experiments in addition to algorithm analysis; it isbelieved that this method is applicable; there is error in bandwidth; this paper proposedan revised method, that is to increase the measurement data packet size and reduce thenumber of load data packets; experiment results show that the revised method hasimproved measurement accuracy. Fourthly, the paper analyzes the network structuresand network performance characteristics of IB network and Ethernet respectively,introduces Qperf, the tool for measuring IB network, and Tperf, the tool for measuring Ethernet, and proves that the tools can achieve effective measurement according tocorresponding network connectivity, bandwidth and delay. Fifthly, based on the studyof the network performance measurement of Internet, IB network and Ethernet as wellas the improvement and the selection of measurement tools, this paper designs anetwork performance measurement system, with detailed description of the overallarchitecture of the system and functions of the models. The effects of the models areproven by experiments.