Internet Draft Network Working Group Kenya Takashima Internet Draft Expiration Date Koji Nakamichi Toshio Soumiya Fujitsu Laboratories Ltd. October 1999 Concept of IP Traffic Engineering draft-takashima-te-concept-00.txt Status of this Memo This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet- Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt. The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. Abstract There is an strong demand from IP network operators to obtain a method to control the network resource and enhance network performance. The answer for this is Traffic Engineering. It is derived through the effort in MPLS wg. But current, the concept of Traffic Engineering seems to be limited to a few keywords, namely load balancing and restoration. The aim of this document is to enhance the concept of traffic engineering and point out some aspects of what it is capable of. The Takashima, et al. [Page 1] Internet Draft draft-takashima-te-concept-00.txt October1999 goal of this document is accomplished if this document could enlighten the perspective of IP Traffic Engineering and show a possible direction. 1. Introduction World has paid great attention to the Internet for accommodating not only data communication service like web service, but also real time streaming service like voice and video. With the increasing importance of Internet as the commercial communication infrastructure, many requirements has been imposed to the network. Growth of the traffic forces the operational network to make efficient use of the network resource. And the newer applications request the network to provide some kind of QoS. This situation has triggered the need for Traffic Engineering. Making use of Traffic Engineering, network is able to achieve higher utilization of the resource in a intelligent manner. However, at this point in time, the word of Traffic Engineering points two limited keywords, network load balancing and network restoration. The aim of this document is to enhance the concept of traffic engineering and point out some aspects of what it is capable of. As described in section 3., the concept of IP Traffic Engineering is to optimize the network performance by controlling the network. In other words, IP Traffic Engineering should provides the functionality to realize the optimized solution for how the operation be done when serving QoS service and best effort service in a single network. The remainder of this draft is organized as follows. Section 2 discusses background. Section 3 discusses the concept of traffic engineering. Section 4 discusses the performance objective of the Traffic Engineering. And Section 5 describes what the requirements from the applications would be and discusses what IP Traffic Engineering should focus on. 2. Background Internet has been paid a great attention for accommodating not only data communication service like web, but also real time streaming service like voice and video. Now Internet has becoming a common infrastructure for multimedia communications. When building communication infrastructure over the Internet, method for engineering the network will be the important factor. Network engineering has been studied using the telecommunication traffic theory for many years, and it has supported the evolution and the Takashima, et al. [Page 2] Internet Draft draft-takashima-te-concept-00.txt October1999 knowledge of maintaining the current telecommunication networks. It is necessary for the Internet as a commercial infrastructure to adopt such engineering method. Essentially, some control function is needed when delivering data from source to destination in the network. There are two types of control method, micro-level control and macro-level control. The micro-level control is a function of equipment in network and is a control at the flow level. The flow level control includes admission control, policing control, quality control and congestion control. These set of control is also known as traffic control. On the other hand, the macro-level control is the network wide control, and it resolves a problem that micro-level control can not. For both micro and macro-level controls, it could be either static control or dynamic control, and the operation could be either concentrated or distributed. In the current Internet, complicated traffic control functions were moved from network to terminal side in order to simplify the network. This is TCP. TCP is categorized in the window-based congestion control of the micro-level control and works at end-to-end. Basically, it was considered that network should not interfere with TCP. What the network could do in the Internet was to just provide fat pipes. In other words, ISPs could only depended on a primitive method as follows; manually construct new physical link when the operator detects with his eyes that the utilization of an active link exceeded a threshold. However, as the importance of Internet as the communication infrastructure increases, many problems which can not be solved using conventional methods have appeared. Especially, the following issues are important; - QoS assurance - Traffic Engineering The former is caused by the increase of services like video streaming. Some control function is needed for network to assure the end-to-end QoS because TCP can only assure the packet delivery. The latter is the issue of the following; (1) congestion can not be controlled even if the physical links are added, (2) network resource can not use effectively even if the number of multipath or route increases. Due to the paradigm of the conventional IP routing, the resource can not be allocated efficiently. Current network can not be engineered as operators wishes. The concept of Traffic Engineering is described in section 3. Traffic engineering has the objective to optimize all of the network resource. To be more direct, Traffic Engineering optimizes the network performance by balancing QoS and network utilization of non-QoS resource, where two elements are controversial to each other. Takashima, et al. [Page 3] Internet Draft draft-takashima-te-concept-00.txt October1999 It is done by taking into account the link utilization and/or QoS. General view is that the cause of the above issue is originated in IP routing protocol. So it is likely that Traffic Engineering should take into account of some kind of QoS aware routing protocol as a tool. Sorting out general functions of Traffic Engineering is a future issue. Traffic Engineering is macro-level control, but it is different from network engineering which uses telecommunication traffic theory. Traffic Engineering does not depend on a specific layer 2 technology. However, from the viewpoint of implementation, currently MPLS is the most suitable tool. MPLS has provided some core concepts for Traffic Engineering, i.e. FEC and traffic trunk, and developed protocols to set up explicit routes. In a sense, it can be said that MPLS has opened a new paradigm, a technology category called IP Traffic Engineering. 3. Concept of Traffic Engineering Traffic Engineering provides automatic optimization of network resource in order to satisfy performance objective of application service, and it is done by taking into account the performance and/or QoS that can be provided by the network. Deciding the number of equipments and the configuration of the network that would satisfy the given performance, when developing, designing, and maintaining a network. The network engineering of conventional telecommunication has been done based on communication traffic theory, which has been used for expanding and maintaining current telecommunication networks. Internet is completely different from the telecommunication network, but some engineering method is necessary in order to use the Internet as the common communication infrastructure. IP Traffic Engineering will play an important role in the operational IP network. Traffic Engineering provides a flexible way of responding to the demand which changes day by day, and it optimizes the network to satisfy the required performance from the network operator. The aim of Traffic Engineering is to achieve a network that is "Network Engineering Free". Up until this time, IP network was intended to provide only reachability. But recent applications show demands for higher quality. For example, when a path for QoS service is set up between ingress and egress in a domain, load balancing with optimization function is applied. When the QoS guaranteed path become active in a link that both best effort traffic and QoS service traffic is present, Traffic Engineering enables the best effort service without degrading service availability, say, using a function that reroute Takashima, et al. [Page 4] Internet Draft draft-takashima-te-concept-00.txt October1999 the best effort path to the other link. This is a control which satisfy a request for guaranteeing QoS and high link utilization, which are controversial to each other. The balancing parameter is the operator's policy. The ultimate view of how the Traffic Engineering would operate might be as follows: network operator sets the parameters for defined per service profile, and Traffic Engineering function interpret it to achieve the network optimization in a totally automatic manner. 4. Traffic Engineering Performance Objectives As stated in previous section, the goal of traffic engineering is to avoid the congestion problem and the degradation of QoS caused by those congestion, and to do it automatically in a so called "network engineering free" manner. From the viewpoint of traffic in the network, the function of the Traffic Engineering could be taken as to optimize the distribution of the traffic in the network. When considering the performance objectives of traffic engineering, we can classified them into following two points according to from which side the traffic is viewed in the network: - Application Oriented Performance Objective, - Network Oriented Performance Objective 4.1 Application Oriented Performance Objective Application oriented performance objective is an objective related to traffic characteristics of the individual flow corresponds to each application or service. This is QoS related objective and attempt to improve following characteristics: - end-to-end packet transmission delay, - packet delay variation, - server response time, etc. End-to-end packet transmission delay is transferring time between end users or edge nodes. Packet delay variation is the variation of packet inter-arrival time at a given node. These parameters have great impact for real-time base application such as VoIP. Server response time is the round trip delay until user sending packet returns back to the user again. These parameters related to the performances of client-server type application such as web hosting. Takashima, et al. [Page 5] Internet Draft draft-takashima-te-concept-00.txt October1999 Below explains application oriented performance objective by taking load balancing as an example. Load balancing is one of the key function of traffic engineering. Generally, IGP(Interior Gateway Protocol) routing in the Internet determines only a single route bound to an given destination. In this case, we can easily imagine that such a single route causes congestion because a number of flows concentrate on this route and packets of the flows rush into the same route. As a result, this degrades above characteristics. Delay sensitive applications are fatal under such congestion. Load balancing, however, which splits traffic flows which bound to same egress router into multiple routes and distributes traffic loads, release congestion for such flows and reduce the probability that encounters any other congestion. As a result, packets can be transferred without any degradation of traffic characteristics and requirements from application can be satisfied. 4.2 Network Oriented Performance Objective Network oriented performance objective is an objective related to the network resources. This attempt to improve following characteristics: - network resource utilization, - throughput, etc. Network resource utilization is a ratio of total traffic amount to the network total bandwidth. This indicates an efficiency of network resource usage. Throughput is a traffic transfer rate for whole network at an given time. This indicates a traffic transferring capacity in whole network. Traffic engineering expects high efficiency of network resource. Consider the previous load balancing. As already pointed out above, as long as followed the single route determined by IGP routing, the flows passes the route with high probability. In this sense, IGP's single route makes traffic in the network localize in this route. Conversely, this means that there are some available resources other than the route in the network. Actually, if there is no flows other than that between said ingress and egress, the route except the route with flow are empty. Load balancing mechanism splits incoming flows at ingress and into multiple routes so that localized traffic can be distributed into whole network. This could lead to high network efficiency and high throughput. 4.3 Traffic Engineering Performance Optimization Traffic engineering have effects on both traffic oriented and resource oriented performance objectives. In the above examples, Takashima, et al. [Page 6] Internet Draft draft-takashima-te-concept-00.txt October1999 load balancing mechanism can provide improvement of QoS of each flow and network resource efficiency, but not both at same time. To keep the QoS characteristic, such as end-to-end packet transfer delay, some amount of bandwidth must be prepare and result in low bandwidth utilization. In a network, it is not possible to satisfy both application oriented performance objective and network oriented performance objective simultaneously. They are of a trade-off relation. The goal of traffic engineering is to optimize the both performance objectives and balance them. In the viewpoint of traffic in the network, traffic engineering balances QoS traffic and best effort traffic in the network. The decision which objective has higher priority should depend on each network operator's policy. Framework which reflects such an policy would be needed in the future. 5. traffic engineering requirements At this point in time, the word Traffic Engineering only points two things, network load balancing and network restoration. Load balancing averages the load between links and as a result the network resource is optimized. Network restoration provides network robustness and stability. This draft enhances the concept of traffic engineering and give a new view of what IP Traffic Engineering could do. As described in section 3., the concept of IP Traffic Engineering is to optimize the network performance by controlling the network. In other words IP Traffic Engineering provides the functionality to realize the policies which deals with network performance. QoS routing could almost be taken as a example of IP Traffic Engineering. Only thing is that it is poorly engineered. QoS routing hunts for a route to match the requested QoS. But it does not take into account the optimization of the network performance. This document defines that the behavior of optimizing these two factors, providing QoS and sustaining total network performance, is the goal of Traffic Engineering. For example, suppose that QoS is provided by pre-empting the resource from the best effort network, when end-to-end QoS request is induced to the network, IP Traffic Engineering function hunts for a route that meets the QoS which give the least impact to the best effort traffic. By optimizing the network as such, it provides QoS and at the same time keeps the performance of the best effort network. The meaning of optimization of the network performance varies, since the performance objective depends on IP application. Thus, the development of IP Traffic Engineering would start by providing solutions per IP application and gradually merging each result to reach generalized form of IP Traffic Engineering. In subsection 5.1, example requirements from the existing IP application is shown. In subsection 5.2, a form of future IP traffic engineering is observed. Takashima, et al. [Page 7] Internet Draft draft-takashima-te-concept-00.txt October1999 5.1 requirements from applications At the beginning of section 5, is was described that the objective of IP Traffic Engineering should be set from the requirements which derives from each IP application. That is, the goal of IP Traffic Engineering differs depending on IP applications. Recent IP applications apt to require at least some form of quality. This quality might be QoS parameters, or could be user sensed quality. In this section, a few example of requirements from IP application is shown. In addition, some indication of how IP Traffic Engineering would function with the requirements is briefly described. traffic engineered web service Web hosting provides WWW contents service and is one of the most important IP service. Part of the requirements Web hosting requests is shown below. - high accessability to the server - quick response time - assuring QoS parameter(bandwidth reservation, 100% connectivity) for streaming data - fast database synchronization between mirror servers without inducing impact to the network QoS(bandwidth reservation, delay bound) could be achieved by means of layer2 technology, and as for more loose quality request such as accessability, improved response time, robustness of the service could be achieved by explicit routing with algorithms which is a technology to optimize the performance of the network. Another perspective of web hosting with Traffic Engineering could be found from the database point of view. The key factors for WWW contents service are where to put the database and how to transfer them. By taking into account the geographical placement of servers, usage of the cache system, this would help optimize the web server infrastructure and could result in improved web service. Currently, server load balancing product is in the market. This provides local load balancing between servers. Network wide load balancing is done by using NAT. Introducing IP Traffic Engineering would enhance the effect of load balancing and provides method to achieve other requirements from the application. Takashima, et al. [Page 8] Internet Draft draft-takashima-te-concept-00.txt October1999 traffic engineered multicast service Requirements from multicast service includes delay guaranty, jitter guaranty, controlling constraint in number of channels, constraint in network utilization. When multicast service is engineered, one of the key factor would be optimization between multicast traffic and unicast traffic. This optimization may be done by rerouting unicast traffic from the links which mulitcast tree is using. traffic engineered VoIP Requirements from IP telephony includes low call denial probability, delay assurance, jitter assurance, assurance to preserve connection. Also ISP would serve voice traffic with high priority because of its high income rate. Important factor in VoIP Traffic Engineering is to control the effect of call acceptance to best effort service and optimize the performance. 5.2 future requirements What we understand from section 5.1 is that the performance objective of the Traffic Engineering is application specific. The objective for one application may be providing least hop paths and for other application the objective may be reserving bandwidth. When the knowledge of operating the service is sufficiently provided, the relationship between service quality and the cost will become clear. Since IP network is undoubtedly becoming the infrastructure for every communication system, the knowledge of simultaneous operation of various applications or services. In other words, the technique to know how an operation of one service affects others and to control the behavior in order to optimize the network performance. This aspect implies that optimization of the network resource as a whole is required. An example could be taken from operation of VoIP service. When the network is serving VoIP at busy hours, a service provider would likely use extra network resource for this particular service so as to sustain the quality(call acceptance ratio, bandwidth per call). Under this condition, it is likely that the quality of the web hosting will deteriorate. In this case, to find out which link to be pre-empted for providing extra resource to VoIP is the most effectless in performance to Web hosting is the objective. Optimization of the operational network in this fashion is a high importance and IP Traffic Engineering is the function to accomplish this need. Note that it may be efficient to make use of the policy framework to articulate all of these IP Traffic Engineering objectives. Takashima, et al. [Page 9] Internet Draft draft-takashima-te-concept-00.txt October1999 Great portion of the current IP application takes form of server-client model. Client receives a service by accessing the database at the server. From IP Traffic Engineering point of view, the two important factors of this model are where to place the database and how to transport the data. Once this general optimization solution is obtained, network operation is given a step to be released from the geographical constraints. To put this broadly, IP Traffic Engineering will lead to a engineering free operational network. 6. routing consideration For further study. 7. network scope of traffic engineering The scope of the network which traffic engineering is applied depends on the application. It could be intra-domain or could be among several ISPs. The effect of traffic engineering is fully obtained only when the method for inter-domain engineering is established. 8. conclusion This draft presents a concept of what IP Traffic Engineering should take into account. The elements emphasized are as follows: - the objective of the IP Traffic Engineering is to optimize the network performance - network performance optimization is obtained by providing QoS and sustaining total network performance of non QoS environment. - the objective of the IP Traffic Engineering should meet the requirements from various applications 9. references [1] D.Awduche et al., Requirements for Traffic Engineering Over MPLS, RFC 2702, September 1999 [2] D.Awduche et al., Extensions to RSVP for LSP Tunnels, draft-ietf-mpls-rsvp-lsp-tunnel-04.txt, September 1999 [3] B.Jamoussi et al., Constraint-Based LSP Setup using LDP, draft-ietf-mpls-cr-ldp-03.txt, September 1999 [4] R. Callon et al., A Framework for Multiprotocol Label Switching, draft-ietf-mpls-framework-05.txt, September 1999 Takashima, et al. [Page 10] Internet Draft draft-takashima-te-concept-00.txt October1999 10. authors' addresses Kenya Takashima Fujitsu Laboratories Ltd. 4-1-1 Kamikodanaka, Nakahara-ku, Kawasaki 211-8588, Japan E-mail: kenya@flab.fujitsu.co.jp Koji Nakamichi Fujitsu Laboratories Ltd. 4-1-1 Kamikodanaka, Nakahara-ku, Kawasaki 211-8588, Japan E-mail: nakamichi@flab.fujitsu.co.jp Toshio Soumiya Fujitsu Laboratories Ltd. 4-1-1 Kamikodanaka, Nakahara-ku, Kawasaki 211-8588, Japan E-mail: soumiya@flab.fujitsu.co.jp Takashima, et al. [Page 11]