Network Performance Perception in the Framework of NGN (tutorial proposal for APNOMS 2007)

Transcription

1 Network Performance Perception in the Framework of NGN (tutorial proposal for APNOMS 2007) Marat Zhanikeev Waseda University, Japan 1. A two-sentence description End-to-end network performance is gaining increasing interest in NGN standardization. Various views on network performance and methods of its passive or active measurement are discussed in this tutorial. 2. A two-paragraph description of the tutorial NGN is a wide term that strives to deliver various services existing in separate technological planes today over an all-ip network, i.e. using packet-switching only. Naturally, in such a network, various contents, such as video, voice, and text will have to coexist regardless of differences in QoS requirements made by each of them separately. NGN deals with this boost in complexity by separating control place from the transport plane. Services, therefore, will be defined and delivered at the control plane, while transport layer will be used for transport-only. Currently, ITU defines 6 distinct QoS classes for IP networks in Y.1541 recommendation based basic network characteristics, such as mean and statistical upper bound of transfer delay and packet loss, etc. These characteristics, however, define only the transport network, while application QoS requirements defined in G.1010 prove to be much richer and require a non-trivial mapping to be performed between these two definitions of QoS. Since the above deals with the general area of network performance, it is important to define network performance based on various ways existing today to perceive it through passive and active measurement. This tutorial discusses passive measurements based on RMON MIBs and active measurements targeting end-to-end performance metrics defined by IETF IPPM in the framework of heterogenous services of NGN. 3. A detailed outline of the tutorial NGN separates control place from transport place in the new network design. Transport plane will be composed of access and core IP networks that will be used to provide global connectivity in all-ip networks, both wired and wireless. Control plane, sometimes referred to as service layer, is to be used to connect services and is defined in an abstract way so that services would not depend on underlying transport network technology. The main task of NGN, however, is to move all currently existing non-ip network technologies to packet switching. The move is complicated by various QoS requirements on the part of various existing technologies. It is difficult for traditional packet switching to support fine QoS granularity. Therefore, traditional disadvantages of all-ip networks apply to NGN networks. This includes generally unpredictable multimodal probability distribution of traffic in IP networks. This is a major change from traditional telephone networks, where arrival rate has a static value. Traffic in IP networks at very long time intervals follows multi-fractal distribution, both the mean and variance of which are unknown. This imposes major limitations on definition of end-to-end transfer delay and packet loss included in Y.1541 recommendation on QoS classes in NGN.

2 Network performance has only recently become a part of NGN standardization process. First, Y.1541 defines 5 distinct QoS classes based on simple performance metrics, such as end-to-end delay and packet loss. Delay variation were also included in Y.1541 in form 99.9% guarantee that the variation (jitter) will not exceed a certain value. Some QoS classes have relaxed entries, where either end-to-end transfer delay or jitter are marked as U for unpredictable, which accommodates best-effort nature of today s IP networks. On the other hand, applications perceive network performance in more detail than transport layer offers. Instead of end-to-end delay and jitter alone, some real-time applications perceive network performance in terms of traffic bursts, which are described by maximum achievable throughput, average/maximum burst length, and other similar characteristics. Generally, QoS definitions at application layer may seem less stringent but are much more diverse in terms of performance characteristics. This discrepancy in perception of network performance by applications and NGN transport layer is yet to be addressed by NGN standardization process in several years to come. Before an abstract mathematical view of network performance can be created, it is necessary to review the basics of queuing theory as applied to packet-switched networks. Some traditional teletraffic theory will also be reviewed lightly. Finally, based on queueing and teletraffic theory, some network performance models will be considered. Also, since NGN is predicted to make extensive use of optical networks and MPLS technology, network performance in regard to these two relatively new technologies will be discussed. Passive measurement will constitute a substantial part of the tutorial as the earliest technologies created for network performance measurement. SNMP protocol and several most popular MIBs will be considered, including the most popular RMON MIB, which was included in NGN standardization in 2006 as a means for collection of performance measurement data and was referred to as Performance Measurement Management. Naturally, limitations of MIB-based passive performance measurements will be discussed in relation of end-to-end nature of NGN services. Finally, active measurement technology will be discussed. Sometimes referred to as probing, active measurements are methods that use packet probes to measure network performance in an active way. Since probing is easier to be performed in end-to-end manner, active measurements are perfectly fit for performance measurement in the framework of NGN. Basic probing technology and its mathematical apparatus will be displayed and followed by a number of examples of discovering network performance characteristics by active probing. Some main performance characteristics to be discussed are bulk transfer capacity, available bandwidth, throughput, etc. As many active measurement tools exist today, the tutorial will give an overview on the most popular active measurement suites, such as NetPerf, IPerf and others. Some benchmarking targeting TCP throughput performance will also be included and considered from the viewpoint of network-centric applications. Some of active measurement tools and multi-purpose active measurement software were created and previously published by the authors and will also be presented at the tutorial. In the conclusion of the tutorial, two separate models of performance measurements within NGN specifications will be considered. One model will confine measurement tasks within transport layer only and will consider creation of a flexible platform, which will maintain performance information obtained from measurements among multiple Measurement Points (MP) in a large-scope transport network. Abstract service layer will be able to acces performance data through the API defined for transport layer by NGN. Since, many MPs will be participating in such a large-scale measurement infrastructure, the issues of complexity and scalability will be the two primary issues considered under this model. The second model will consider inclusion of end-to-end measurements in SIP agents. SIP agents are defined by IMS to replace traditional phone networks in transition to all-ip networks. However,

3 as was recently proved by active research, the use of SIP could go beyond traditional audio-video point-to-point communications, and could be used for delivery of rich multimedia content. SIP was defined in a relaxed manner, allowing the protocol to work seamlessly over best-effort portions of the network, and use only the switches that have SIP functionality included in them. This is also convenient for end-to-end measurements, which do not need to be included in every switch on a path. Active measurement tasks conducted over end-to-end SIP paths will be considered in the tutorial as the second model for network performance measurement within NGN. The preliminary table of contents of the tutorial is given below: 1. NGN Framework 2. QoS in NGN 2.1 Transport-layer QoS requirements, performance classes 2.2 Application QoS requirements 2.3 Making transport and application QoS requirements meet 3. Network Performance Basics 3.1 Queuing theory 3.2 Dependence of performance on network architecture 3.3 Performance of multi-hop paths 3.4 Aggregated performance of a network 4. Passive measurement 4.1 SNMP 4.2 RMON: Remote monitoring MIB 4.3 Billing 5. Active measurement 5.1 IPPM: End-to-end performance metrics 5.2 End-to-end measurement methods 5.3 Existing end-to-end measurement tools 5.4 Advanced active measurement targets 5.5 Applications and active measurement results 6. Performance Measurement Players in NGN 6.1 Active methods for performance measurement role 6.2 Passive methods for performance measurement management role 6.3 Performance measurement in MPLS networks 6.4 Performance measurement in optical networks 6.5 Measurements using multiple MP in NGN transport layer 6.6 Measurements using end-to-end SIP paths. 6.7 Some other practical measurement cases 7. Characterization of the potential target audience and prerequisite knowledge People participating in standardization of performance measurement in NGN and those working in the area of network performance monitoring and measurement in general should constitute the majority of the audience. However, the content of the tutorial can interest researchers from a wide range of topics, from next generation network applications to new network architectures and new protocols. Since prerequisite knowledge is limited to the very basic notions of packet communications and TCP/IP protocol stack, the contents of the tutorial will be comprehensive for researchers and engineers from areas other than network performance. 8. Reasons why the tutorial topic should be or interest to APNOMS audience QoS and Network Performance (the terms can be used interchangeably) have only recently been

4 included in NGN standardization process. This already happened before when TCP/IP emerged and proved to be unable to provide QoS regardless of congestion control features included into the protocol itself. This was the main reason why network performance measurement became a ground for active research discussion. When applications came forth and demanded their version of QoS, another major change in performance measurement was triggered and resulted in a wide variety of measurement tools that exist today. Today, active measurements are part of any real-time or QoS-enabled network application. NGN standardization is still in the early stage of development, where QoS was defined only in very abstract terms of end-to-end transfer delay, jitter and loss. IETF IPPM (IP Performance Metrics), however, are already included in NGN QoS specifications to play the role of Performance Measurement along with RMON MIBs playing the role of Performance Measurement Management. This specification, however, is still very vague, and will require to clear many details before QoS at application layer can be implemented in practice. This tutorial gives an extensive and detailed overview of performance measurement technology existing today in best-effort networks and can be a good starting point from which to define performance measurement technology of NGN. 9. Brief resume of the presenter and contact details Marat Zhanikeev received the B.S. degree in electrical and electronics engineering from Tashkent State Technical University, Tashkent, Uzbekistan, and M.S. and PhD in information and telecommunications studies from Waseda University, Tokyo, Japan, in 1997, 2003, and 2007, respectively. From 1997 to 2001, he had been working as an engineer in Daewoo Telecom Tashkent on a number of telecommunications modernization projects within the governmental development program. From 2004 to 2007, he was employed as Research Associate by Waseda University. His current research interests include network measurement, network monitoring, and network management. He is a Regular Member of IEICE. Contact details: Marat Zhanikeev, Yoshiaki Tanaka RISE, Bldg.41, Room 303, Kikuicho 17 Shinjuku-ku, Tokyo JAPAN maratishe@aoni.waseda.jp tel: Major publications by the presenter related to the contents of the tutorial Master Thesis (1) M.Zhanikeev, A Study on Dynamic Bottleneck Bandwidth Measurements, Global Information and Telecommunications Studies, Waseda University,July PhD Dissertation Journal Publications (2) M.Zhanikeev, A Study on Active Measurement in the Internet," Global Information and Telecommunications Studies, Waseda University,February (3) M.Zhanikeev and Y.Tanaka, Temporal Patterns and Properties in Multiple-Flow Interactions, Management of Convergence Networks and Services, Lecture Notes in Computer Science (LNCS), Vol.4238, pp , Springer, September (4) W.Tan, M.Zhanikeev, and Y.Tanaka, Rate-Based and Gap-Based Available Bandwidth Estimation Techniques in Cross-Traffic Context, Management of

5 International Conferences Conference Provisioning Convergence Networks and Services, Lecture Notes in Computer Science (LNCS), Vol.4238, pp.73-81, Springer, September (best student paper award) (5) T.Q.Le, M.Zhanikeev, and Y.Tanaka, Detecting and Identifying Network Anomalies by Component Analysis, Management of Convergence Networks and Services, Lecture Notes in Computer Science (LNCS), Vol.4238, pp , Springer, September (6) M.Zhanikeev and Y.Tanaka, Network Management Using Active Probing, 12th International Conference on Telecommunication Systems - Modeling and Analysis (ICTSM12), Monterey, U.S.A., Session 9, pp , July (7) M.Zhanikeev and Y.Tanaka, Network Performance Optimization through Measurement, International Network Optimization Conference (INOC 2005), Lisbon, Portugal, Session MC4, pp.b1.265-b1.272, March (8) M.Zhanikeev and Y.Tanaka, Towards the Improvement of Performance Anomaly Prediction, 1st IEEE and IFIP International Conference in Central Asia on Internet (ICI 2005), Bishkek, Kyrgyz Republic, Session 7, 5 pages, September (9) M.Zhanikeev, J.McKeown, and Y.Tanaka, Data Sources for Proactive Network Management, 8th Asia-Pacific Network Operations and Management Symposium (APNOMS 2005), Okinawa, Japan, Session 5, pp , September (10) M.Zhanikeev and Y.Tanaka, A Testbed for Agent-Based Multi-Purpose Extensible Active Measurement, 2nd International IEEE/Create-Net Conference on Testbeds and Research Infrastructures for the Development of Networks and Communities (TridentCom 2006), Barcelona, Spain, Track 2 - Session 4, 9 pages, March (11) M.Zhanikeev, S.Xu, and Y.Tanaka, Active Performance Measurement for IP over All-Optical Networks, 2nd IEEE and IFIP International Conference in Central Asia on Internet (ICI 2006), Tashkent, Uzbekistan, Session 7, 5 pages, September (12) M.Zhanikeev and Y.Tanaka, Temporal Patterns in Size and Rate of Traffic Flows, 12th International Telecommunications Network Strategy and Planning Symposium (Networks 2006), New Delhi, India, Technical Session 4, 6 pages, November (13) W.Tan, M.Zhanikeev, and Y.Tanaka, ABshoot: A Reliable and Efficient Scheme for End-to-End Available Bandwidth Measurement, IEEE TENCON 2006 (Region 10 Conference), Hong Kong, China, Paper No.DN3.3, 4 pages, November (14) T.Q.Le, M.Zhanikev, and Y.Tanaka, Anomaly Identification Based on Flow Analysis, IEEE TENCON 2006 (Region 10 Conference), Hong Kong, China, Paper No.SC1.3, 4 pages, November (15) T.Q.Le, M.Zhanikev, and Y.Tanaka, Methods of Distinguishing Flash Crowds from Spoofed DoS Attacks, 3rd EURO-NGI Conference on Next Generation Internet Networks, Trondheim, Norway, May (16) Secretary, 2nd IEEE and IFIP International Conference in Central Asia on Internet (ICI 2006), September 2005 September (17) Secretary, 3rd IEEE and IFIP International Conference in Central Asia on Internet (ICI 2007), September 2006 current.