| Network Working Group | F. Brockners |
| Internet-Draft | S. Bhandari |
| Intended status: Informational | S. Dara |
| Expires: September 14, 2017 | C. Pignataro |
| Cisco | |
| H. Gredler | |
| RtBrick Inc. | |
| J. Leddy | |
| Comcast | |
| S. Youell | |
| JMPC | |
| D. Mozes | |
| Mellanox Technologies Ltd. | |
| T. Mizrahi | |
| Marvell | |
| P. Lapukhov | |
| R. Chang | |
| Barefoot Networks | |
| March 13, 2017 |
Requirements for In-situ OAM
draft-brockners-inband-oam-requirements-03
This document discusses the motivation and requirements for including specific operational and telemetry information into data packets while the data packet traverses a path between two points in the network. This method is referred to as "in-situ" Operations, Administration, and Maintenance (OAM), given that the OAM information is carried with the data packets as opposed to in "out-of-band" packets dedicated to OAM. In situ OAM complements other OAM mechanisms which use dedicated probe packets to convey OAM information.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
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."
This Internet-Draft will expire on September 14, 2017.
Copyright (c) 2017 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
This document discusses requirements for "in-situ" Operations, Administration, and Maintenance (OAM) mechanisms. In this context, “in-situ OAM” refers to the concept of directly encoding telemetry information within the data packet as it traverses the network or telemetry domain. Mechanisms which add tracing or other types of telemetry information to the regular data traffic, sometimes also referred to as "in-band" OAM can complement active, probe-based mechanisms such as ping or traceroute, which are sometimes considered as "out-of-band", because the messages are transported independently from regular data traffic. In terms of "active" or "passive" OAM, "in-situ" OAM can be considered a hybrid OAM type. While no extra packets are sent, in-situ OAM adds information to the packets therefore cannot be considered passive. In terms of the classification given in [RFC7799] in-situ OAM could be portrayed as "hybrid OAM, type 1". "In-situ" mechanisms do not require extra packets to be sent and hence don't change the packet traffic mix within the network. Traceroute and ping for example use ICMP messages: New packets are injected to get tracing information. Those add to the number of messages in a network, which already might be highly loaded or suffering performance issues for a particular path or traffic type.
A number of in-situ as well as in-band OAM mechanisms have been discussed, such as the INT spec for the P4 programming language [P4] or the SPUD prototype [I-D.hildebrand-spud-prototype]. The SPUD prototype uses a similar logic that allows network devices on the path between endpoints to participate explicitly in the tube outside the end-to-end context. Even the IPv4 route-record option defined in [RFC0791] can be considered an in-situ OAM mechanism. Per what was already stated, in-situ OAM complements "out-of-band" mechanisms such as ping or traceroute, or more recent active probing mechanisms, as described in [I-D.lapukhov-dataplane-probe]. In-situ OAM mechanisms can be leveraged where current out-of-band mechanisms do not apply or do not offer the desired characteristics or requirements, such as proving that a certain set of traffic takes a pre-defined path, strict congruency between overlay and underlay transports is in place, checking service level agreements for the live data traffic, detailed statistics or verification of path selections within a domain, or scenarios where probe traffic is potentially handled differently from regular data traffic by the network devices. [RFC7276] presents an overview of OAM tools.
Compared to probably the most basic example of "in-situ OAM" which is IPv4 route recording [RFC0791], an in-situ OAM approach has the following capabilities:
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119].
Abbreviations used in this document:
This document defines in-situ Operations, Administration, and Maintenance (in-situ OAM), as the subset in which OAM information is carried along with data packets. This is as opposed to "out-of-band OAM", where specific packets are dedicated to carrying OAM information.
In several scenarios it is beneficial to make information about the path a packet took through the network or through a network device as well as associated telemetry information available to the operator. This includes not only tasks like debugging, troubleshooting, as well as network planning and network optimization but also policy or service level agreement compliance checks. This section discusses the motivation to introduce new methods for enhanced in-situ network diagnostics.
Packet scheduling algorithms, especially for balancing traffic across equal cost paths or links, often leverage information contained within the packet, such as protocol number, IP-address or MAC-address. Probe packets would thus either need to be sent from the exact same endpoints with the exact same parameters, or probe packets would need to be artificially constructed as "fake" packets and inserted along the path. Both approaches are often not feasible from an operational perspective, be it that access to the end-system is not feasible, or that the diversity of parameters and associated probe packets to be created is simply too large. An in-situ mechanism is an alternative in those cases.
In-situ mechanisms are not impacted by differences in the handling of probe traffic compared to other data packets, where probe traffic is handled differently (and potentially forwarded differently) by a router than regular data traffic. This obviously assumes that the addition of in-situ information does not change the forwarding behavior of the packet. Note that in certain implementations, the addition information to a transport protocol changes the forwarding behavior. IPv6 extension header processing is one example. Some implementations process IPv6 packets with extension headers in the "slow" path of a router, as opposed to the "fast" path.
Traditional ping and traceroute tools return the OAM results to the sender of the probe. Even when the ICMP messages that are used with these tools are enhanced, and additional telemetry is collected (e.g., ICMP Multi-Part [RFC4884] supporting MPLS information [RFC4950], Interface and Next-Hop Identification [RFC5837], etc.), it would be advantageous to separate the sending of an OAM probe from the receiving of the telemetry data. In this context, it is helpful to eliminate the requirement that there be a working bidirectional path.
Several network deployments leverage tunneling mechanisms to create overlay or service-layer networks. Examples include VXLAN-GPE, GRE, or LISP. One often observed attribute of overlay networks is that they do not offer the user of the overlay any insight into the underlay network. This means that the path that a particular tunneled packet takes, nor other operational details such as the per-hop delay/jitter in the underlay are visible to the user of the overlay network, giving rise to diagnosis and debugging challenges in case of connectivity or performance issues. The scope of OAM tools like ping or traceroute is limited to either the overlay or the underlay which means that the user of the overlay has typically no access to OAM in the underlay, unless specific operational procedures are put in place. With in-situ OAM the operator of the underlay can offer details of the connectivity in the underlay to the user of the overlay. This could include the ability to find out which underlay elements are shared by overlays and ability to know which overlays are mapped to the same underlay elements. Deployment dependent underlay transit nodes can be configured to update OAM information in the overlay transport encapsulation. The operator of the egress tunnel router could choose to share the recorded information about the path with the user of the overlay.
Coupled with mechanisms such as Segment Routing (SR) [I-D.ietf-spring-segment-routing], overlay network and underlay network can be more tightly coupled: The user of the overlay has detailed diagnostic information available in case of failure conditions. The user of the overlay can also use the path recording information as input to traffic steering or traffic engineering mechanisms, to for example achieve path symmetry for the traffic between two endpoints. [I-D.brockners-lisp-sr] is an example for how these methods can be applied to LISP.
In-situ OAM can help users of an overlay-service to verify that negotiated SLAs for the real traffic are met by the underlay network provider. Different from solutions which rely on active probes to test an SLA, in-situ OAM based mechanisms avoid wrong interpretations and "cheating", which can happen if the probe traffic that is used to perform SLA-check is prioritized by the network provider of the underlay. In active/standby deployments in-situ OAM would only allow for SLA verification of the active path.
Network planners and operators benefit from knowledge of the actual traffic distribution in the network. When deriving an overall network connectivity traffic matrix one typically needs to correlate data gathered from each individual device in the network. If the path of a packet is recorded while the packet is forwarded, the entire path that a packet took through the network is available to the egress system. This obviates the need to retrieve individual traffic statistics from every device in the network and correlate those statistics, or employ other mechanisms such as leveraging traffic engineering with null-bandwidth tunnels just to retrieve the appropriate statistics to generate the traffic matrix.
In addition, with individual path tracing, information is available at packet level granularity, rather than only at aggregate level - as is usually the case with IPFIX-style methods which employ flow-filters at the network elements. Data-center networks which use equal-cost multipath (ECMP) forwarding are one example where detailed statistics on flow distribution in the network are highly desired. If a network supports ECMP, one can create detailed statistics for the different paths packets take through the network at the egress system, without a need to correlate/aggregate statistics from every router in the system. Transit devices are off-loaded from the task of gathering packet statistics.
In high-speed networks one can leverage and benefit from packet-accurate measurements with for example hardware-accurate timestamping (i.e., nanosecond-level verification) to support optimized packet scheduling and queuing mechanisms.
Bandwidth- and power-constrained, time-sensitive, or loss-intolerant networks (e.g., networks for industry automation/control, health care) require efficient OAM methods to decide when to replicate packets to a secondary path in order to keep the loss/error-rate for the receiver at a tolerable level - and also when to stop replication and eliminate the redundant flow. Many Internet of Things (IoT) networks are time sensitive and cannot leverage automatic retransmission requests (ARQ) to cope with transmission errors or lost packets. Transmitting the data over multiple disparate paths (often called bi-casting or live-live) is a method used to reduce the error rate observed by the receiver. Time sensitive networks (TSN) receive a lot of attention from the manufacturing industry as shown by a various standardization activities and industry forums being formed (see e.g., IETF 6TiSCH, IEEE P802.1CB, AVnu).
Several deployments use traffic engineering, policy routing, segment routing or Service Function Chaining (SFC) [RFC7665] to steer packets through a specific set of nodes. In certain cases regulatory obligations or a compliance policy require to prove that all packets that are supposed to follow a specific path are indeed being forwarded across the exact set of nodes specified. If a packet flow is supposed to go through a series of service functions or network nodes, it has to be proven that all packets of the flow actually went through the service chain or collection of nodes specified by the policy. In case the packets of a flow weren't appropriately processed, a verification device would be required to identify the policy violation and take corresponding actions (e.g., drop or redirect the packet, send an alert etc.) corresponding to the policy. In today's deployments, the proof that a packet traversed a particular service chain is typically delivered in an indirect way: Service appliances and network forwarding are in different trust domains. Physical hand-off-points are defined between these trust domains (i.e., physical interfaces). Or in other terms, in the "network forwarding domain" things are wired up in a way that traffic is delivered to the ingress interface of a service appliance and received back from an egress interface of a service appliance. This "wiring" is verified and trusted. The evolution to Network Function Virtualization (NFV) and modern service chaining concepts (using technologies such as Locator/ID Separation Protocol (LISP), Network Service Header (NSH), Segment Routing (SR), etc.) blurs the line between the different trust domains, because the hand-off-points are no longer clearly defined physical interfaces, but are virtual interfaces. Because of that very reason, networks operators require that different trust layers not to be mixed in the same device. For an NFV scenario a different proof is required. Offering a proof that a packet traversed a specific set of service functions would allow network operators to move away from the above described indirect methods of proving that a service chain is in place for a particular application.
Deployed service chains without the presence of a "proof of transit" mechanism are typically operated as fail-open system: The packets that arrive at the end of a service chain are processed. Adding "proof of transit" capabilities to a service chain allows an operator to turn a fail-open system into a fail-close system, i.e. packets that did not properly traverse the service chain can be blocked.
A solution approach could be based on OAM data which is added to every packet for achieving Proof Of Transit (POT).The OAM data is updated at every hop and is used to verify whether a packet traversed all required nodes. When the verifier receives each packet, it can validate whether the packet traversed the service chain correctly. The detailed mechanisms used for path verification along with the procedures applied to the OAM data carried in the packet for path verification are beyond the scope of this document. Details are addressed in [I-D.brockners-proof-of-transit]. In this document the term "proof" refers to a discrete set of bits that represents an integer or string carried as OAM data. The OAM data is used to verify whether a packet traversed the nodes it is supposed to traverse.
In-situ OAM could be leveraged for several use cases, including:
The implementation of an in-situ OAM mechanism needs to take several considerations into account, including administrative boundaries, how information is recorded, Maximum Transfer Unit (MTU), Path MTU Discovery (PMTUD) and packet size, etc.
The information gathered for in-situ OAM can be categorized into three main categories: Information with a per-hop scope, such as path tracing; information which applies to a specific set of hops, such as path or service chain verification; information which only applies to the edges of a domain, such as sequence numbers. Note that a single network device could comprise several in-situ OAM hops, for example in case one wants to trace the path of a packet through that device.
The recorded data at every hop might lead to packet size exceeding the Maximum Transmit Unit (MTU). A detailed discussion of the implications of oversized IPv6 header chains is found in [RFC7112]. The Path MTU restricts the amount of data that can be recorded for purpose of OAM within a data packet.
If in-situ OAM data is inserted at the edge of the domain (e.g., by intermediate routers) then the MTU on all interfaces with the domain (MTU_INT) MUST be >= the maximum MTU on any "external" facing interfaces (MTU_EXT) and the total size of in-situ OAM data to be recorded MUST be <= (MTU_INT - MTU_EXT).
In-situ OAM comprises two approaches to insert OAM data fields in the packets:
The "incremental" or the "pre-allocated" approaches could even be combined in the same deployment - in which case two in-situ OAM headers would be present in the packet: One for the incremental approach and one for the pre-allocated approach. In such a case one would expect that nodes with a hardware data-plane would update the incremental header, whereas nodes with a software data-plane would process the pre-allocated header.
There are several challenges in enabling in-situ OAM in the public Internet as well as in corporate/enterprise networks across administrative domains, which include but are not limited to:
The following considerations will be discussed in a future version of this document: If the packet is dropped due to the presence of the in-situ OAM; If the policy failure is treated as feature disablement and any further recording is stopped but the packet itself is not dropped, it may lead to every node in the path to make this policy decision.
Like any OAM domain, in-situ OAM domains could also be layered/nested. Layering/nesting of in-situ OAM follows the general approach of OAM layering: An in-situ OAM domain consists of maintenance end-points (MEP) and maintenance intermediate points (MIP). MEP add to or remove the entire set of in-situ OAM data fields from the traffic, while only MIP update or add in-situ OAM data fields. When in-situ OAM layering is employed, a MEP of one layer becomes a MIP in the layer above, while MIP of the lower layer are not visible to the layer above - unless specifically configured otherwise.
Consider the following examples:
The ability to selectively enable in-situ OAM is valuable. While it may be desirable to enable data collection on all traffic or devices, this may not always be feasible. In-situ OAM collection may also come with a performance impact to forwarding rates or feature capabilities, which may be acceptable in only some locations. For example, the SPUD prototype uses the notion of "pipes" to describe the portion of the traffic that could be subject to in-path inspection. Mechanisms to decide which traffic would be subject to in-situ OAM are outside the scope of this document.
In-situ OAM adds additional data fields to live user traffic and as such changes the packet which is also why in-situ OAM is characterized as "hybrid, type 1" OAM. The effectiveness of in-situ OAM as a tool for operations depends on forwarding nodes not altering their forwarding behavior in case of in-situ OAM data fields being present in the packet. As a consequence, an implementation of in-situ OAM should not change the forwarding behavior of the packet, i.e. packets with or without in-situ OAM data fields should be handled the same way by a forwarding node (see also the associated requirement further below). Note that there are implementations where the addition of meta-data to live user traffic might cause the forwarding behavior of the packet to change, e.g. certain implementation handle IPv6 packets with or without extension headers differently (see [RFC7872]).
Since packets have a finite maximum size, the data recording or carrying capacity of one packet in which the in-situ OAM metadata is present is limited. In-situ OAM should use its own dedicated namespace (confined to the domain in-situ OAM operates in) to represent node and interface IDs to save space in the header. Generic representations of node and interface identifiers which are globally unique (such as a UUID) would consume significantly more bits of in-situ OAM data.
When recorded data is required to be analyzed on a source node that issues a packet and inserts in-situ OAM data, the recorded data needs to be carried back to the source node.
One way to carry the in-situ OAM data back to the source is to utilize an ICMP Echo Request/Reply (ping) or ICMPv6 Echo Request/Reply (ping6) mechanism. In order to run the in-situ OAM mechanism appropriately on the ping/ping6 mechanism, the following two operations should be implemented by the ping/ping6 target node:
The above discussed use cases require different types of in-situ OAM data. This section details requirements for in-situ OAM derived from the discussion above.
General Security considerations will be expanded on in a later version of this document.
In-situ OAM is considered a "per domain" feature, where one or several operators decide on leveraging and configuring in-situ OAM according to their needs. Still operators need to properly secure the in-situ OAM domain to avoid malicious configuration and use, which could include injecting malicious in-situ OAM packets into a domain.
Threat Model: Attacks on the deployments could be due to malicious administrators or accidental misconfiguration resulting in bypassing of certain nodes. The solution approach should meet the following requirements:
[RFC Editor: please remove this section prior to publication.]
This document has no IANA actions.
The authors would like to thank Jen Linkova, LJ Wobker, Eric Vyncke, Nalini Elkins, Srihari Raghavan, Ranganathan T S, Karthik Babu Harichandra Babu, Akshaya Nadahalli, Ignas Bagdonas, LJ Wobker, Erik Nordmark, Vengada Prasad Govindan, and Andrew Yourtchenko for the comments and advice. This document leverages and builds on top of several concepts described in [I-D.kitamura-ipv6-record-route]. The authors would like to acknowledge the work done by the author Hiroshi Kitamura and people involved in writing it.
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997. |