Internet Draft Internet Engineering Task Force R. Guerin/S. Kamat/E. Rosen INTERNET DRAFT IBM/IBM/Cisco 29 July 1997 Setting up Reservations on Explicit Paths using RSVP draft-guerin-expl-path-rsvp-00.txt Status of This Memo This document is an Internet-Draft. 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 not appropriate to use Internet Drafts as reference material, or to cite them other than as a ``working draft'' or ``work in progress.'' To learn the current status of any Internet-Draft, please check the ``1id-abstracts.txt'' listing contained in the internet-drafts Shadow Directories on ds.internic.net (US East Coast), nic.nordu.net (Europe), ftp.isi.edu (US West Coast), or munnari.oz.au (Pacific Rim). Abstract This document presents motivations for extensions to RSVP in order to enable setting up of reservations on explicit routes. The advantages of providing this support are discussed in the context of MPLS and QoS routing. An approach to providing these extensions by means of opaque routing objects in RSVP messages is presented. Guerin, et al. Expires 3 February 1998 [Page i] Internet Draft RSVP on Explicit Paths 29 July 1997 Contents Status of This Memo i Abstract i 1. Introduction 1 2. Bandwidth Reservation for Explicit Route in an MPLS Environment 1 3. QoS Routing with Explicit Routes 3 3.1. QoS path management . . . . . . . . . . . . . . . . . . . 4 3.2. Enforcing high level admission control policies . . . . . 6 4. Mechanism for Reservation Set Up on Explicit Paths 7 4.1. Explicit Route Object . . . . . . . . . . . . . . . . . . 7 4.2. Processing of Explicit Route Objects . . . . . . . . . . 8 5. Conclusions 8 Guerin, et al. Expires 3 February 1998 [Page ii] Internet Draft RSVP on Explicit Paths 29 July 1997 1. Introduction The purpose of this document is to introduce and motivate extensions to RSVP to enable setting up of reservations on explicit routes. Enabling reservations on explicit routes can be useful in several different contexts. In particular, it can be used to ensure that certain flows use a ``label switched'' path as in the MPLS context [CDF+97] or to facilitate the management of QoS paths computed by a QoS capable router as in [GKO+97]. In this document, we describe further these potential benefits, and show how they can be attained with minimal impact to RSVP. It should be pointed out that the focus of this document is on unicast flows as there are many other issues that need to be addressed to consider the use of explicit routes for multicast flows. In the context of unicast flows, explicit routes are to be specified through a new Explicit_Route object in RSVP. This object, like policy objects, is opaque to RSVP which only needs to ensure its delivery to routing. Routing is responsible for processing the Explicit_Route object, and will use the information it contains to construct its response to a Route_Query from RSVP. Sections 2 and 3 motivate the need for explicit route support within RSVP in the context of MPLS and QoS routing respectively. Section 4 describes the specific mechanism of setting up reservations on explicit paths. This includes specification of a format for the Explicit_Route object and the interactions between RSVP and routing in this context. 2. Bandwidth Reservation for Explicit Route in an MPLS Environment Consider the following topology: A---B---C---D | | E-------F Let us suppose that this topology exists in the network of an Internet Service Provider (ISP). We suppose further that node A has an interface to one of the ISP's subscribers, S1, and node B has an interface to a different subscriber, S2. Finally, we suppose that both subscribers are generating packets that are addressed to destinations reachable only through node D. Guerin, et al. Expires 3 February 1998 [Page 1] Internet Draft RSVP on Explicit Paths 29 July 1997 In order to make the best provisioning of its bandwidth, the ISP may decide that such packets from S1 should follow the route A-B-E-F-D, while such packets from S2 should follow the route B-C-D. Further, the ISP may want to reserve resources for each of these "flows", so that it can schedule transmissions along the respective routes in a way which corresponds to whatever agreements the ISP has made to the particular subscribers. Putting this decision into effect in a conventional IP network is extremely difficult, since it requires that two packets going through B, with the same destination, be sent on separate routes. Therefore, ISPs tend to use ATM or Frame Relay networks to provide this level of bandwidth management. ATM and Frame Relay networks also provide the capability to support whatever resources reservations are necessary. MPLS [CDF+97] provides a way for an ISP to obtain this functionality without the need to resort to ATM or Frame Relay. In MPLS, node A can apply a "label" to packets from S1 which must pass through D; node B can apply a label to packets from S2 which must pass through D. When a labeled packet is transmitted, the label is sent along with it. Once a packet is labeled, the forwarding decision is based only on the label, NOT on the contents of the packet header. Thus there is nothing to prevent packet P1 from traveling the A-B-E-F-D path, while packet P2 travels the B-C-D path, even if P1 and P2 happen to have the very same destination address. Of course, MPLS must incorporate some "path setup" procedure whereby paths that differ from the "normal" IP routing can get explicitly set up. MPLS must also incorporate some means of performing resource reservation along these paths. While a resource reservation protocol could be designed exclusively for MPLS, it would seem to make most sense to use RSVP for that purpose; after all, RSVP was designed to be the resource reservation protocol of the internet. This requires some modification of RSVP. As currently specified, there is no way to force an RSVP Path message to follow any path other than the "normal" path to a particular destination. So if a different MPLS path were set up for certain flows, there is currently no way to get the Path message to follow that path. If RSVP control messages could carry opaque objects that are meaningful to routing and RSVP's interface to routing is broadened as in [GKR97] so that RSVP could pass such objects to routing, then this difficulty can be overcome. The Path messages could carry an explicit route object. To determine the next hop for the flow, RSVP would pass the Explicit Route Object (and other opaque objects if present) to routing, which would pass back the identity of the next hop, and a modified Explicit Route Object. This would force the Path Guerin, et al. Expires 3 February 1998 [Page 2] Internet Draft RSVP on Explicit Paths 29 July 1997 message to follow the path of the corresponding MPLS flow and ensure that resources are reserved along the MPLS path. The general ability to carry an opaque routing object in RSVP messages further enables one to combine the setup of an MPLS path with resource reservation along the same path. This could be achieved by having a second opaque routing object carry an MPLS flow identifier (label) in conjunction with the explicit route object. Clearly, this approach has the advantage of avoiding a second round trip to make reservations along the MPLS path when the path set up itself must be done first. The need to have a second round trip seems to simply add latency and complexity, without adding any value. 3. QoS Routing with Explicit Routes An objective of QoS routing is to choose for each flow the path that has the best likelihood of meeting the flow's QoS requirements, while still making efficient use of network resources. In order to achieve this goal, QoS routing requires knowledge of network resources availability and of the QoS requirements of the flows. This information can be provided in a number of ways (e.g., see [CNRS97, GKR97] for possible approaches) and is then used by a QoS path selection algorithm to identify an appropriate path for a flow. The selection of a path and the distribution of the information needed to make that selection, however, only represent a subset of the functions needed to support QoS routing. There are two other important issues that a QoS routing solution must address to meet its objectives. These are: 1. Management of QoS paths of individual flows, and 2. Enforcing high level admission control policies. Management of QoS paths includes not only setting up the paths correctly, but also maintaining or adjusting them in response to failures and changes in the network. High level (call) admission policies are needed (see [CNRS97] for a discussion of this issue) to control how selected paths are being used so as to preserve the long term efficiency of the network. For example, a suitable path might be found for a flow, but rejected by the high level admission control because of its cost to the network, e.g., it is using a large number of links which could alternatively be used to support several such other calls to different destinations. In the rest of this section, we articulate how explicit routes can facilitate handling of these two issues. However, before doing so, we briefly compare, in the context of QoS routing, the use of Guerin, et al. Expires 3 February 1998 [Page 3] Internet Draft RSVP on Explicit Paths 29 July 1997 explicit routes versus the hop-by-hop routing approach presented in [GKO+97]. A hop-by-hop routing solution has the benefits of requiring the least changes to RSVP and possibly offering added flexibility (see [GKH97] for details), but this does come at a cost. Specifically, with hop-by-hop routing, there are multiple decision points (each hop) involved in selecting a path, with each making independent decisions. As a result, end-to-end control of a path requires coordination between the multiple decision points, and this can often be a complex task. For example, even in the context of a link state routing protocol such as OSPF where all routers in a domain compute their routes using the same algorithm applied to a common topology database, no single router has complete knowledge of the actual path being followed. This is because inconsistencies during routing transients as well as equal cost multi-path considerations, independently affect local path selection decisions. Additional mechanisms are, therefore, needed to coordinate these independent decisions. On the other hand, when explicit routes are used, selection of the entire path is made at a single decision point (the first router in the path). In the rest of this section, we expand on the benefits of a single decision point in the context of both QoS path management and high level call admission. 3.1. QoS path management In best-effort routing, route changes occur relatively infrequently, and mostly when local interfaces change state or when routing updates are received from the routing protocol. With QoS routing, changes that would result in the selection of a new route for a given destination and QoS requirements are much more frequent, as they typically occur each time a metric update is received. If such changes were to trigger re-routing of existing QoS flows, this would translate into disruption of service to already established flows. Furthermore, this could also increase routing instability as such re-routing may trigger additional metric updates and cause further re-routing. Keeping a flow's routing state, i.e., the path on which it has established a reservation, ``pinned'' as long as the path remains satisfactory for the flow (and the network) is one possible approach to this problem. Path pinning, however, has a number of implications for QoS routing. First, path pinning requires knowledge that the path being pinned is adequate. This includes several aspects. First and foremost, the pinned path should be loop free. When an explicit route is used, Guerin, et al. Expires 3 February 1998 [Page 4] Internet Draft RSVP on Explicit Paths 29 July 1997 this is readily achieved as the node selecting the explicit route can ensure it is free of any loop. In contrast, when hop-by-hop routing is used, the coordination of the multiple decision points involved in the selection of the path requires not only that all nodes rely on the same routing algorithm, but also imposes close coupling with RSVP states to detect the formation of loops (see [GKH97] for details). Such a coupling adds some complexity, but more importantly, it can prevent the flow of data on the pinned path until after resources have been successfully reserved on the entire path (see again [ref] for details). In the case where reservations are successful on only a portion of the path, this means that the data may not be able to take advantage of such partial reservations. This is obviously undesirable, and while this can possibly be remedied (see also [GKH97] for possible approaches), solutions come at the cost of added signaling and processing complexity. Besides being loop free, a pinned path must also be capable of satisfying the QoS requirements of the flow. Hence, it is important either to ensure the availability of resources on a pinned path, or to provide simple mechanisms to unpin it in case the required resources are not available when they are being requested, e.g., when an RSVP RESV message is received. Hence, the ability to detect such conditions and trigger the unpinning of a path is required. This can be achieved using similar mechanisms in both the explicit route and hop-by-hop routing cases, and is further detailed in the next section which expands on the approach of [GKH97]. Note that unpinning a path only implies that QoS routing is being queried anew to determine if the current path is still the correct one, or to find if a new and better one now exists. In particular, unpinning a path does not result in removal of existing path or reservation states. This is because although the existing pinned path may not fully satisfy the requirements of the flow, it may be the best one currently available. In that case any (partial) reservation that may exist on the current path should be maintained as it represents the best possible QoS available to the flow. There are other instances where a path needs to be unpinned. For example, when one of the links or nodes on the path fails. In such cases, it is important to notify all nodes on the current path, so that they can unpin it and query QoS routing to possibly find an alternate path. This can again be achieved using similar mechanisms in both the explicit route and hop-by-hop routing cases [GKH97]. However, when a reservation is already in place, it is also desirable to identify links on which resources are already reserved for the flow. This is important so that these existing reservations be taken into account when searching for an alternate path, i.e., avoid the ``stepping on one's shadow'' problem. This is made easier in the case of explicit route by the knowledge of the entire path. Guerin, et al. Expires 3 February 1998 [Page 5] Internet Draft RSVP on Explicit Paths 29 July 1997 Knowledge of the entire path is also useful in the context of high level admission control, and we now briefly review this issue and the benefits of explicit routes in that context. 3.2. Enforcing high level admission control policies As pointed out in the framework document for QoS routing [CNRS97], some form of higher level admission control and administrative control of routing behavior may be necessary within an AS. This is because QoS routing has to balance the sometimes conflicting requirements of high network resource utilization and improved chances of successful resource reservation for individual flows. For example, when current load in the network suggests a QoS path that is much longer than the ``usual'' path, admitting the flow along such a path may actually deny service to later flows that would have been admissible along segments of this long path. Hence, this could negatively affect the overall network utilization. In such situations, a high level admission control policy may find it desirable not to admit the flow based on routing decision alone. One possible approach is to compare the length of the path returned by QoS routing to that of a ``usual'' path, and decide whether or not to use the path depending on this comparison as well as possibly other factors such as overall network load. Conversely, if a flow has been already set up and later a much more efficient path becomes available, it might be desirable to reroute the flow along the new path. This is particularly true if the current path only supports a fraction of the desired reservation, while the new path may be able to accommodate the complete reservation. In all such instances, these decisions are greatly facilitated when a single entity is responsible for determining and controlling the entire path. Hence, such controls are more readily performed when routing is done using explicit routes instead of hop-by-hop routing. This is not to say that they are not feasible with hop-by-hop routing, but distributed decisions and knowledge generally complicate such tasks. For example, transient inconsistent routing information at multiple routers can lead to the pinning of a long but loop-free path, without any single router on the path being aware of the problem. Hence, it becomes difficult to identify and rectify such bad routing choices. Solutions to this problem require the introduction of additional signaling information to coordinate information and decisions across the nodes on the path, e.g., a policy object carried in PATH messages that specifies a limit on the acceptable path length. This would in turn add to the overall signalling and processing overhead, and may all but eliminate the potentially greater simplicity of hop-by-hop routing. On the other Guerin, et al. Expires 3 February 1998 [Page 6] Internet Draft RSVP on Explicit Paths 29 July 1997 hand, the single decision point of explicit routes avoids most of these problems. 4. Mechanism for Reservation Set Up on Explicit Paths 4.1. Explicit Route Object As stated earlier, explicit routes are to be specified through a new Explicit_Route object in RSVP. RSVP PATH messages will carry this object. The format of the Explicit_Route object as shown below consists of a (strict/loose) source route and a pointer indicating the current position within the source route. +-------------+-------------+-------------+-------------+ | Code | Length | Pointer | | +-------------+-------------+-------------+-------------+ | IPv4 Hop 0 (src) Address (4 bytes) | +-------------+-------------+-------------+-------------+ // // +-------------+-------------+-------------+-------------+ | IPv4 Hop N (dst) Address (4 bytes) | +-------------+-------------+-------------+-------------+ This is a new object class, with the C-type indicating IPv4. The value of the Code field will indicate whether the sequence of IP addresses represents a loose or a strict source route. Semantics of Explicit Route Object correspond to the semantics of the IP Strict or Loose Source Route Option as specified in RFC 1812, "Requirements for IP Version 4 Routers". As in the case of processing of IP header source route option processing, routing at intermediate gateways will overwrite the items in the address list with their local network addresses, in effect recording the actual traversed route. This facilitates avoidance of loops which may arise in the case of a loose source route. As described in Section 2 when MPLS is enabled, it may also be desirable to combine the setup of an MPLS with the reservation of resources along that path. This can be achieved through the use of other opaque routing objects, that will carry the MPLS flow information required for the associated path setup. Details on the formats of those object are deferred to other documents dedicated to the setup of MPLS flows. Guerin, et al. Expires 3 February 1998 [Page 7] Internet Draft RSVP on Explicit Paths 29 July 1997 4.2. Processing of Explicit Route Objects Certain additional processing is involved when dealing with explicit route objects as compared to processing a similar source route option in an IP datagram. First, in case the source route object suggests more than one possible outgoing interfaces, routing is free to choose one from among them by any local means. However, it must remember which one it chose, and make the same choice in response to subsequent route queries from RSVP with the same explicit route object. Second, in the event of errors in processing the route object, in stead of generating ICMP messages as in the case of IP datagram processing, routing will notify RSVP of an error. This would trigger a PathErr message to the source with a new error code ("Cannot satisfy source route"). During the processing of the PATH message, RSVP will extract any routing objects (if present) and pass them to routing. Routing will parse these objects implementing their associated semantics and return new opaque objects to replace the original ones in the outgoing messages. As specified in [GKR97] this is achieved by means of Route_Query and Route_Response interfaces: - Route_Query interface is of the form: Route_Query( flow_id, Network header, Transport Header, notify_flag, sender_TSPEC, ADSPEC, opaque_object1, opaque_object2, ...) - Route_Reply is of the form: Route_Reply(flow_id, notify_flag, outgoing_interface_mask, opaque_object1, opaque_object2, ...) 5. Conclusions This document provides a motivation for supporting opaque routing objects in RSVP to enable setting up resource reservations on explicit routes. The benefits of this approach in the contexts of MPLS and QoS routing were expounded and a mechanism for supporting this feature was discussed. References [CDF+97] R. Callon, P. Doolan, N. Feldman, A. Fredette, G. Swallow, and A. Viswanathan. A Framework for Multi-Protocol Label Switching (draft-ietf-mpls-framework-00.txt). INTERNET-DRAFT, Internet Engineering Task Force, May 1997. Guerin, et al. Expires 3 February 1998 [Page 8] Internet Draft RSVP on Explicit Paths 29 July 1997 [CNRS97] E. Crawley, R. Nair, B. Rajagopalan, and H. Sandick. A Framework for QoS-based Routing in the Internet (draft-ietf-qosr-framework-00.txt). INTERNET-DRAFT, Internet Engineering Task Force, March 1997. [GKH97] R. Guerin, S. Kamat, and S. Herzog. QoS Path Management with RSVP, (draft-guerin-qos-path-mgt-rsvp-00.txt). INTERNET-DRAFT, Internet Engineering Task Force, March 1997. [GKO+97] R. Guerin, S. Kamat, A. Orda, T. Przygienda, and D. Williams. QoS Routing Mechanisms and OSPF Extensions, (draft-guerin-qos-routing-ospf-01.txt). INTERNET-DRAFT, Internet Engineering Task Force, March 1997. [GKR97] R. Guerin, S. Kamat, and E. Rosen. Extended RSVP-Routing Interface (draft-guerin-ext-rsvp-rtng-intf-00.txt). INTERNET-DRAFT, Internet Engineering Task Force, July 1997. Authors' Address Roch Guerin IBM T.J. Watson Research Center P.O. Box 704 Yorktown Heights, NY 10598 EMAIL: guerin@watson.ibm.com VOICE +1 914 784-7038 FAX +1 914 784-6205 Sanjay Kamat IBM T.J. Watson Research Center P.O. Box 704 Yorktown Heights, NY 10598 EMAIL: sanjay@watson.ibm.com VOICE +1 914 784-7402 FAX +1 914 784-6205 Eric Rosen Cisco Systems, Inc. 250 Apollo Drive Chelmsford, MA, 01824 EMAIL: erosen@cisco.com Guerin, et al. Expires 3 February 1998 [Page 9]