DiffServ B. Gaidioz Internet-Draft UCBL/ENS-Lyon Expires: August 23, 2002 P. Primet INRIA/ENS-Lyon G. Montenegro Sun Microsystems, Inc. February 22, 2002 The Equivalent Differentiated Services draft-gaudioz-equivalent-00 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. This Internet-Draft will expire on August 23, 2002. Copyright Notice Copyright (C) The Internet Society (2002). All Rights Reserved. Abstract This document describes EDS (Equivalent Differentiated Services), a new building-block for a simple, robust, free and scalable end-to-end service differentiation in IP networks. The EDS scheme aims at providing a spectrum of "different but equivalent" network services that offer a trade-off between delay and loss rate to the end-to-end flows. The EDS scheme can be deployed incrementally in the Internet. Gaidioz, et al. Expires August 23, 2002 [Page 1] Internet-Draft Equivalent Differentiated Services February 2002 1. Introduction and Requirements With the diversification of Internet applications, flow sensitivity to loss or delay variations becomes increasingly heterogeneous. This highlights an important need for service differentiation at the IP layer. As compared to best-effort service, some flows require better delay and others, better loss rate. For example, whereas audio streams are very sensitive to delay variations, different types of TCP flows exhibit highly varying reponses to network conditions, depending on whether they are "bulky" (e.g. large file transfers over extended periods of time) or "interactive" (e.g. telnet with short interactive packets). Furthermore, the web has introduced flows which are both bulky and interactive (interactive web pages with large multimedia content like pictures, audio or video clips). Accordingly, an important challenge is how to complement the traditional best-effort service by adding the appropriate differentiation mechanisms. In keeping with the Internet's design principles, these mechanisms must adhere to the following criteria [3], [4]: o They must be simple, robust and scalable, o They must be incrementally deployable. o The various services must be usable as freely as the current best- effort service provided by today's IP networks. Neither pricing nor admission control must be performed. Active queue management mechanisms such as RED [9] and congestion notification mechanisms like ECN [8] aim at improving the best-effort service by having detecting congestion early and issuing notifications to responsive flows. Active queue management fulfills the above criteria. Nevertheless, providing differentiated services is not one of their objectives. To avoid pricing, the proposed services cannot constitute a hierarchy from "better" to "worse". Consequently, even if flows do obtain different treatment by the network, what a flow gains in one respect, it must loose in another. This is the underlying principle of "non- elevated services" [5] like ABE [1] or BEDS [2]. Admission control can be avoided only if guarantees are relative rather than absolute. Absolute guarantees requires resource provisioning, admission control and traffic control to ensure traffic is policed accordingly. If guarantees are relative, they can still be ensured, whatever the network load is. Gaidioz, et al. Expires August 23, 2002 [Page 2] Internet-Draft Equivalent Differentiated Services February 2002 The requirement for simplicity rules out the potential treatment of individual packets. Besides, performance at the network layer must high and compatible with the current and evolving link speeds. Today's routers are able to classify packets and to treat them differently in different queues efficiently. But, with improvements in optical technology and the exponential increase of the network capacity, the complexity of packet processing performed by routers at ultra high speeds cannot increase much more. The robustness requirement implies that the network remains stateless at the core in order to support services in an end-to-end fashion. Incremental deployment means that the scheme must support service differentiation even if it is performed only partially along the end- to-end path. EDS is a new building block at the IP layer that satisfies the preceding criteria. It aims at differentiating the per hop behavior (PHB) to better meet the needs of each flow. EDS provides a spectrum of different but equivalent services, that offer a trade-off between delay and loss rate. Gaidioz, et al. Expires August 23, 2002 [Page 3] Internet-Draft Equivalent Differentiated Services February 2002 2. Specifications The EDS defines an arbitrary number N of equivalent service classes (N greater or equal to 2) which are identified by numbers ranging from `1' to `N'. The services are directly used by end-to-end protocols or applications. Each "EDS-capable router" differentiates among classes over the queuing delay of the packets and their loss rate. A class `i' is given two constant coefficients `d_i' (the delay coefficient) and `l_i' (the loss rate coefficient) defined as follows. Let `i' and `j' be two different classes: in each router, a ratio of d_i/d_j between the queuing delays of their packet and a ratio of l_i/l_j between their loss rates are defined. Coefficients are set so that for each i in [1,N-1], l_i+1 is higher than l_i and d_i+1 is lower than d_i. Class `1' is thus the class whose packets experience the lowest loss rate and the highest queuing delay ; packets of class `N' gets the highest loss rate and the lowest queuing delay. Packets of classes `i' where 1