Hsiaosu Hsiung Senior Principal Engineer Mitretek Systems Inc 7525 Colshire Drive McLean VA 22102 hhsiung@mitretel.org (v) (703) 610-2931 (f) (703) 610-2984 Quality Of Service Performance Measurement INTRODUCTION Much research devoted to the Next Generation Internet (NGI) will focus on developing data applications requiring real-time, high-speed, continuous data flows to support fully interactive video, audio, multimedia, and enhanced virtual reality applications used in scientific research, national security, medicine, environmental monitoring, distance education, and health care. In order to guarantee the network performance quality required of these applications, research is focusing on quality of service (QoS) -based routing and its associated network performance management techniques for real-time flow measurement. This paper will discuss performance management architecture that are candidate for research initiative. Each of the following six interrelated elements offer a framework for providing network performance management. 7 Performance measurement architecture 7 Performance metrics 7 Network performance measurement tools 7 Data collection 7 Data analysis 7 Network performance optimization The performance measurement architecture is defined in conjunction with the appropriate metrics. Then, measurement tools that are employed in the network s devices (e.g., SNMP agents) collect, organize, and classify the collected protocol data (e.g., L2 and L3 data). Results from the analysis of the collected data are used to adjust the operational characteristics in optimizing the network s operation to meet the QoS requirements. The same results can be used to optimize the measurement architecture and the associated performance metrics as well. PERFORMANCE MEASUREMENT ARCHITECTURE The selection of the performance measurement architecture is critical to accommodating the amount of traffic and the number and proximity of the network nodes involved. The architecture must be adaptable and scaleable to support a diverse set of performance measurements for current and future needs. The complex networks supporting the NGI require decentralized, distributed measurement architectures. In a distributed architecture, performance is measured by network management agents within network devices. The data collected by these agents is processed, filtered, and forwarded to a centralized network station. Appropriate network management protocols must be selected to enable adequate observation and secure control capabilities. CMIP and SNMP, with extensions, are being investigated along with the applicability of the MIB II data formats. Security issues regarding confidentiality of acquired data and protection of network control from unauthorized sources are also being addressed. PERFORMANCE METRICS In light of the performance measurement architecture implemented, the industry needs to agree upon the correct performance metrics to characterize the quality of a service. After identifying the metrics, industry should determine the proper mappings from these metrics to specific measurable parameters. Extensive work is underway now by groups, such as the IETF Quality of Service Routing Working Group and the ATM Forum. Nevertheless, acceptable threshold values remian to be quantified. Sufficient empirical data collected from test and evaluation networks, both government and commercial, should be acquired to relate acceptable network application performance to acceptable network device performance thresholds. NETWORK PERFORMANCE MEASUREMENT TOOLS Development of measurement tools will be closely coordinated with the type of measurement architecture and the performance metrics defined. Any tools developed will be integrated with the network devices. These tools should be platform independent (e.g., work equally well on Intel or RISC based platforms). DATA COLLECTION Data collection will be conducted either actively or passively. Active collection should not be disruptive to existing network operation. Passive collection will acquire data based on only existing network traffic. To obtain empirical performance, data should be collected from live networks, such as the industry supported 6Bone and government supported IPv6 networks. Data collection methods will address how best to handle security issues, including the management of private and confidential information. DATA ANALYSIS Data analysis applies the acquired data to algorithms. These algorithms use acquired data to compile statistical metrics, which then can be used to characterize the network performance. Development of these algorithms is essential for deriving automated characterization of service quality. Simulations may be conducted using NGI-class networks operated in a laboratory environment. However, more realistic results would be obtained from testing done on live networks. In any event, the analysis effort should be accomplished in conjunction with the network performance optimization. NETWORK PERFORMANCE OPTIMIZATION One of the end results of applying network performance management is the optimization of network application performance. Therefore, the results of the data analysis will be used to improve the operational characteristics of the network devices to modify their operation and, consequently, the associated network traffic flows, thus resulting in improved QoS. The amount and type of control exercised over the network devices will be determined by the algorithms developed during data analysis. Note that unauthorized control of the network devices can be prevented by implementing security controls. CONCLUSION The next generation of Internet standards and products are currently being developed by the IETF committees and industry. To ensure successful network operation, they also need to define the network performance parameters that will enable network measurement, characterization, and management functions as discussed in this paper. Furthermore, the methodologies for analyzing the measurement results need to be developed, especially for use during the transitional phase to the NGI. Finally, to validate the analysis findings, live data from a controlled operating network of sufficient size is critical. Such networks suitable to this purpose usually are found in Government- related project environments.