Re: [Detnet] [6tisch] comments on draft-finn-detnet-architecture-00

[Qin Wang <[e-mail]>]

Hi Patrick,
I don't understand why communication jitter = 0 is so important. Assume DetNet can guarantee the maximum end-to-end latency, and there is a well designed buffer in the Listener device, then I think the Listener can consume data just on time, in another word, from application point of view, jitter =0. Please point out if I'm wrong.
ThanksQin 


     On Thursday, April 9, 2015 5:03 PM, Patrick Wetterwald (pwetterw) <[e-mail]> wrote:
   

 Norm, Pascal,
Some other comments:
   
   - If you look at https://tools.ietf.org/html/draft-wetterwald-detnet-utilities-reqs you will see that for some important use cases, the jitter should very close to zero. In the architecture draft, you are mainly focussing on the maximum value of the latency and the packet loss (at least in the text). This is not sufficient for power automation and tele protection use case. We need to be able to deploy deterministic network with Jitter = 0.
   - Second: may be good to reference the requirement drafts in the doc. :-)
Thanks,
Patrick

From: Thomas Watteyne <[e-mail]>
Date: Monday 23 March 2015 18:44
To: "[e-mail]" <[e-mail]>, "[e-mail]" <[e-mail]>
Subject: [Detnet] comments on draft-finn-detnet-architecture-00


Norm, Pascal,
Please find below a number of comment on draft-finn-detnet-architecture-00.
Overall, the draft is very well written and highlights important points for deterministic networking.
My general comments are:- in the context of a low-power wireless network (e.g. 6TiSCH), I wonder what you thoughts are on reserving a fixed path. Maybe some extra text on redundancy is needed?- would it make sense to discuss the difference between hard and soft real-time, also wrt wireless systems?
Thomas
---

DetNet                                                           N. FinnInternet-Draft                                                P. ThubertIntended status: Standards Track                                   CiscoExpires: September 10, 2015                                March 9, 2015

                 Deterministic Networking Architecture                   draft-finn-detnet-architecture-00
Abstract
   Deterministic Networking (DetNet) provides a capability to carry   specified unicast or multicast data streams for real-time   applications with extremely low data loss rates and maximum latency.TW> it looks like you aim for maximum latency :)   Techniques used include: 1) reserving data plane resources for   individual (or aggregated) DetNet streams in some or all of the relay   systems (bridges or routers) along the path of the stream; 2)   providing fixed paths for DetNet streams that do not rapidly change   with the network topology; and 3) sequentializing, replicating, and   eliminating duplicate packets at various points to ensure the   availability of at least one path.  The capabilities can be managed   by configuration, or by manual or automatic network management.
Status of This Memo
   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 10, 2015.
Copyright Notice
   Copyright (c) 2015 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


Finn & Thubert         Expires September 10, 2015               [Page 1]Internet-Draft    Deterministic Networking Architecture       March 2015

   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.
Table of Contents
   1.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . .   2   2.  Terminology . . . . . . . . . . . . . . . . . . . . . . . . .   4   3.  Providing the DetNet Quality of Service . . . . . . . . . . .   5     3.1.  Zero Congestion Loss  . . . . . . . . . . . . . . . . . .   6     3.2.  Pinned-down paths . . . . . . . . . . . . . . . . . . . .   7     3.3.  Seamless Redundancy . . . . . . . . . . . . . . . . . . .   7   4.  DetNet Architecture . . . . . . . . . . . . . . . . . . . . .   8     4.1.  The Application Plane . . . . . . . . . . . . . . . . . .  11     4.2.  The Controller Plane  . . . . . . . . . . . . . . . . . .  11     4.3.  The Network Plane . . . . . . . . . . . . . . . . . . . .  12     4.4.  Elements of DetNet Architecture . . . . . . . . . . . . .  13     4.5.  DetNet streams  . . . . . . . . . . . . . . . . . . . . .  14       4.5.1.  Talker guarantees . . . . . . . . . . . . . . . . . .  14       4.5.2.  Incomplete Networks . . . . . . . . . . . . . . . . .  15     4.6.  Data Flow Model through Systems . . . . . . . . . . . . .  16     4.7.  Queuing, Shaping, Scheduling, and Preemption  . . . . . .  16     4.8.  Coexistence with normal traffic . . . . . . . . . . . . .  16     4.9.  Fault Mitigation  . . . . . . . . . . . . . . . . . . . .  16     4.10. Protocol Stack Model  . . . . . . . . . . . . . . . . . .  17     4.11. Advertising resources, capabilities and adjacencies . . .  17     4.12. Provisioning model  . . . . . . . . . . . . . . . . . . .  17       4.12.1.  Centralized Path Computation and Installation  . . .  17       4.12.2.  Distributed Path Setup . . . . . . . . . . . . . . .  17   5.  Related IETF work . . . . . . . . . . . . . . . . . . . . . .  18     5.1.  Deterministic PHB . . . . . . . . . . . . . . . . . . . .  18     5.2.  6TiSCH  . . . . . . . . . . . . . . . . . . . . . . . . .  18   6.  Security Considerations . . . . . . . . . . . . . . . . . . .  19   7.  IANA Considerations . . . . . . . . . . . . . . . . . . . . .  19   8.  Acknowledgements  . . . . . . . . . . . . . . . . . . . . . .  19   9.  Informative References  . . . . . . . . . . . . . . . . . . .  19   Authors' Addresses  . . . . . . . . . . . . . . . . . . . . . . .  23
1.  Introduction
   Operational Technology (OT) refers to industrial networks that are   typically used for monitoring systems and supporting control loops,   as well as movement detection systems for use in process control   (i.e., process manufacturing) and factory automation (i.e., discrete   manufacturing).  Due to its different goals, OT has evolved in


Finn & Thubert         Expires September 10, 2015               [Page 2]Internet-Draft    Deterministic Networking Architecture       March 2015

   parallel but in a manner that is radically different from IT/ICT,TW> define IT and ICT at first use   focusing on highly secure, reliable and deterministic networks, with   limited scalability over a bounded area.
   The convergence of IT and OT technologies, also called the Industrial   Internet, represents a major evolution for both sides.  The work has   already started; in particular, the industrial automation space has   been developing a number of Ethernet-based replacements for existing   digital control systems, often not packet-based (fieldbus   technologies).
   These replacements are meant to provide similar behavior as the   incumbent protocols, and their common focus is to transportTW> "on transporting"?   a fully   characterized flow over a well-controlled environment (i.e., a   factory floor), with a bounded latency, extraordinarily low frame   loss, and a very narrow jitter.  Examples of such protocols include   PROFINET, ODVA Ethernet/IP, and EtherCAT.TW> It might be good to quantify the target latency, packet loss and jitter.TW> I appreciate this is not easy, but one representatve example might help   In parallel, the need for determinism in professional and home audio/   video markets drove the formation of the Audio/Video Bridging (AVB)   standards effort of IEEE 802.1.  With the explosion of demand for   connectivity and multimedia in transportation in general, the   Ethernet AVB technology has become one of the hottest topics, in   particular in the automotive connectivity.  It is finding application   in all elements of the vehicleTW> ":"   from head units, to rear seat   entertainment modules, to amplifiers and camera modules.  While aimed   at less critical applications than some industrial networks, AVB   networks share the requirement for extremely low packet loss rates   and ensured finite latency and jitter.TW> ensure -> guarantee?
   Other instances of in-vehicle deterministic networks have arisen as   well for control networks in cars, trains and buses, as well as   avionics, with, for instance, the mission-critical "Avionics Full-   Duplex Switched Ethernet" (AFDX) that was designed as part of the   ARINC 664 standards.  Existing automotive control networks such as   the LIN, CAN andTW> would TTP/C fit in this list?   FlexRay standards were not designed to cover these   increasing demands in terms of bandwidth and scalability that we see   with various kinds of Driver Assistance Systems (DAS) and new   multiplexing technologies based on Ethernet are now getting traction.
   The generalization of the needs for more deterministic networks have   led to the IEEE 802.1 AVB Task Group becoming the Time-Sensitive   Networking (TSN) Task Group (TG), with a much-expanded constituency   from the industrial and vehicular markets.  Along with this   expansion, the networks in consideration are becoming larger and   structured, requiring deterministic forwarding beyond the LAN   boundaries.  For instance, Industrial Automation segregates the   network along the broad lines of the Purdue Enterprise Reference


Finn & Thubert         Expires September 10, 2015               [Page 3]Internet-Draft    Deterministic Networking Architecture       March 2015

   Architecture (PERA), using different technologies at each level, and   public infrastructures such as Electricity Automation require   deterministic properties over the Wide Area.  The realization is now   coming that the convergence of IT and OT networks requires Layer-3,   as well as Layer-2, capabilities.
   The present architecture is the result of a collaboration of the IETF   and the IEEE and implements an abstract model that can be applicable   both at Layer-2 and Layer-3, and along segments of different   technologie.TW> typo   With this new work, a path may span, for instance,   across a (limited) number of 802.1 bridges and then a (limited)   number of IP routers.TW> add somethind like "possibly wireless" somewhere?   In that example, the IEEE 802.1 bridges may be   operating at Layer-2 over Ethernet whereas the IP routers may be   6TiSCH nodes operating at Layer-2 and/or Layer-3 over the IEEE   802.15.4e MAC.
   Many applications of interest to Deterministic Networking require the   ability to synchronize the clocks in end systems to a sub-microsecond   accuracy.TW> why this accuracy?   Some of the queue control techniques defined in   Section 4.7 also require time synchronization among relay systems.   The means used to achieve time synchronization are not addressed in   this document.
2.  Terminology
   The follwingTW> typo   special terms are used in this document in order to   avoid the assumption that a given element in the archetectureTW> typo   does or   does not have Internet Protocol stack, functions as a router or a   bridge, or otherwise plays a particular role at Layer-3 or higher:TW> I don't get the sentence above
   bridge           A Customer Bridge as defined by IEEE 802.1Q           [IEEE802.1Q-2011].
   circuit           A trail of configuration from talker to listener(s)TW> sender and receiver might be more common at the IETF?           through           relay systems associated with a DetNet stream, required to           deliver the benefits of DetNet.
   end system           Commonly called a "host" in IETF documents, and an "end           station" is IEEE 802 documents.  End systems of interest to           this document are talkers and listeners.
   listener           An end system capable of sinking a DetNet stream.
   relay system


Finn & Thubert         Expires September 10, 2015               [Page 4]Internet-Draft    Deterministic Networking Architecture       March 2015

           A router or a bridge.           TW> maybe add "layer 2 bridge" and "layer 3 router" for clarity?   
   stream           A DetNet stream is a sequence of packets from a single           talker, through some number of relay systems to one or more           listeners, that is limited by the talker in its maximum           packet size and transmission rate, and can thus be ensured           the DetNet Quality of Service (QoS) from the network.
   talker           An end system capable of sourcing a DetNet stream.
3.  Providing the DetNet Quality of Service
   DetNet Quality of Service is expressed in terms of:
   o  Minimum and maximum end-to-end latency from talker to listener;
   o  Probability of loss of a packet, assuming the normal operation of      the relay systems and links;
   o  Probability of loss of a packet in the event of the failure of a      relay system or link.
   It is a distinction of DetNet that it is concerned solely with worst-   case values for all of the above parameters.  Average, mean, or   typical values are of no interest, because they do not affect the   ability of a real-time system to perform its tasks.TW> Would it make sense to discuss hard and soft real-time here, inTW> particular in the context of a wireless system?   
   Three techniques are employed by DetNet to achieve these QoS   parameters:
   a.  Zero congestion loss (Section 3.1).  Network resources such as       link bandwidth, buffers, queues, shapers, and scheduled input/TW> add 6TiSCH cells?       output slots are assigned in each relay system to the use of a       specific DetNet stream or group of streams.  Note that, given a       finite amount of buffer space)TW> extra ")"   , zero congestion loss necessarily       ensures a maximum end-to-end latency.  Depending on the method       employed, a minimum latency can also be achieved.
   b.  Pinned-down paths (Section 3.2).  Point-to-point paths or point-       to-multipoint trees through the network from a talker to one or       more listeners can be established, and DetNet streams assigned to       follow a particular path or tree.TW> Using a single path in a wireless system will fail, I would recommendTW> discussing here path redundancy, for example through DAGs
   c.  Packet replication and deletion (Section 3.3).  End systems and/       or relay systems can sequence number, replicate, and eliminate       replicated packets at multiple points in the network in order to


Finn & Thubert         Expires September 10, 2015               [Page 5]Internet-Draft    Deterministic Networking Architecture       March 2015

       ensure that one (or more) equipment failure events still leave at       least one path intact for a DetNet stream.
   These three techniques can be applied independently, giving eight   possible combinations, including none (no DetNet), although some   combinations are of wider utility than others.  This separation keeps   the protocol stack coherent and maximizes interoperability with   existing and developing standards in this (IETF) and other Standards   Development Organizations.  Some examples of typical expected   combinations:
   o  Pinned-down paths (a) plus packet replication (b) are exactly the      techniques employed by [HSR-PRP].  Pinned-down paths are achieve      by limiting the physical topology of the network, and the      sequentialization, replication, and duplicate elimination      facilitated by packet tags added at the front or the end of      Ethernet frames.
   o  Zero congestion loss (a) alone is is offered by IEEE 802.1 Audio      Video bridging [IEEE802.1BA-2011].  As long as the network suffers      no failures, near-zero (at best, zero) congestion loss can be      achieved through the use of a reservation protocol (MSRP) and      shapers in every relay system (bridge).
   o  Using all three together gives maximum protection.
   There are, of course, simpler methods available (and employed, today)   to achieve levels of latency and packet loss that are satisfactory   for many applications.  However, these methods generally work best in   the absence of any significant amount of non-critical traffic in the   network (if, indeed, such traffic is supported at all), or work only   if the critical traffic constitutes only a small portion of the   network's theoretical capacity, or work only if all systems are   functioning properly, or in the absence of actions by end systems   that disrupt the network's operations.
   There are any number of methods in use, defined, or in progress for   accomplishing each of the above techniques.  It is expected that this   DetNet Architecture will assist various vendors, users, and/or   "vertical" Standards Development Organizations (dedicated to a single   industry) to make selections among the available means of   implementing DetNet networks.
3.1.  Zero Congestion Loss
   The primary means by which DetNet achieves its QoS assurances is to   completely eliminate congestion at an output port as a cause of   packet loss.  Given that a DetNet stream cannot be throttled, this


Finn & Thubert         Expires September 10, 2015               [Page 6]Internet-Draft    Deterministic Networking Architecture       March 2015

   can be achieved only by the provision of sufficient buffer storage at   each hop through the network to ensure that no packets are dropped   due to a lack of buffer storage.TW> and the provision the sufficient link bandwidth?
   Ensuring adequate buffering requires, in turn, that the talker, and   every relay system along the path to the listener (or nearly every   relay system -- see Section 4.5.2) be careful to regulate its output   to not exceed the data rate for any stream, except for brief periosTW> typo   when making up for interfering traffic.  Any packet sent ahead of its   time potentially adds to the number of buffers required by the next   hop, and may thus exceed the resources allocated for a particular   stream.
   The low-level mechanisms described in Section 4.7 provide the   necessary regulation of transmissions by an edge system or relay   system to ensure zero congestion loss.  Of course, the reservation of   the bandwidth and buffers for a stream requires the provisioning   described in Section 4.12.
3.2.  Pinned-down paths
   In networks controlled by typical peer-to-peer protocols such as IEEE   802.1 ISIS bridged networks or ETF OSPF routed networks, a network   topology event in one part of the network can impact, at least   briefly, the delivery of data in parts of the network remote from the   failure or recovery event.  Thus, even redundant paths through a   network, if controlled by the typical peer-to-peer protocols, do not   eliminate the chances of brief losses of contact.  For this reason,   many real-time networks rely on physical rings of two-port devices,   with a relatively simple ring control protocol.  This both minimizes   recovery time and easily supports redundant paths.  Of course, this   comes at the cost of increased hop count, and thus latency, for the   typical path.
   In order to get the advantages of low hop count and still ensure   against even brief losses of connectivity, DetNet employs pinned-down   paths, where the path taken by a given DetNet stream does not change,   at least immediately, and likely not at all, in response to network   topology events.  When combined with seamless redundancy   (Section 3.3), this results in a high likelihood of continuous   connectivity.
3.3.  Seamless Redundancy
   After congestion loss has been eliminated, the most important causes   of packet loss are random media and/or memory faults and equipment   failures.



Finn & Thubert         Expires September 10, 2015               [Page 7]Internet-Draft    Deterministic Networking Architecture       March 2015

   Seamless redundancy involves three capabilities:
   o  Adding sequence numbers to the packets of a DetNet stream.
   o  Replicating these packets and, typically, sending them along at      least two different paths to the listener(s).
   o  Discarding duplicated packets.
   In the simplest case, this amounts to replicating each packet in a   talker that has two interfaces, and conveying them through the   network, along separate paths, to the similarly dual-homed listeners,   that discard the extras.  This ensures that one path (with zero   congestion loss) remains, even if some relay system fails.TW> this has some energy cost associated?      Alternatively, relay systems in the network can provide replication   and elimination facilities at various points in the network, so that   multiple failures can be accommodated.
   This is shown in the following figure, where the two relay systems   each replicate (R) the DetNet stream on input, sending the stream to   both the other relay system and to the end system, and eliminated   duplicates (E) on the output interface to the right-hand end system.   Any one linksTW> typo   in the network can fail, and the Detnet stream can   still get through.  Furthermore, two links can fail, as long as they   are in different segments of the network.
                     > > > > > > > >   relay    > > > > > > > >                    > /------------+ R system E +------------\ >                   > /                  v + ^                 \ >   end    R +                   v | ^                  + E end   system   +                   v | ^                  +   system                   > \                  v + ^                 / >                    > \------------+ R relay  E +------------/ >                     > > > > > > > >   system   > > > > > > > >
                                 Figure 1TW> I don't understand what R and E means
4.  DetNet Architecture
   Traffic Engineering Architecture and Signaling (TEAS) [TEAS] defines   traffic-engineering architectures for generic applicability across   packet and non-packet networks.  From TEAS perspective, Traffic   Engineering (TE) refers to techniques that enable operators to   control how specific traffic flows are treated within their networks.
   Because if its very nature of establishing pinned-down optimized   paths, Deterministic Networking can be seen as a new, specialized


Finn & Thubert         Expires September 10, 2015               [Page 8]Internet-Draft    Deterministic Networking Architecture       March 2015

   branch of Traffic Engineering, and inherits its architecture with a   separation into planes.
   The Deterministic Networking architecture is thus composed of three   planes, a (User) Application Plane, a Controller Plane, and a Network   Plane, which echoes that of Software-Defined Networking (SDN): TW> double ":"   Layers   and Architecture Terminology [RFC7426] which is represented below:











































Finn & Thubert         Expires September 10, 2015               [Page 9]Internet-Draft    Deterministic Networking Architecture       March 2015

           SDN Layers and Architecture Terminology per RFC 7426
                     o--------------------------------o                     |                                |                     | +-------------+   +----------+ |                     | | Application |   |  Service | |                     | +-------------+   +----------+ |                     |       Application Plane        |                     o---------------Y----------------o                                     |       *-----------------------------Y---------------------------------*       |           Network Services Abstraction Layer (NSAL)           |       *------Y------------------------------------------------Y-------*              |                                                |              |               Service Interface                |              |                                                |       o------Y------------------o       o---------------------Y------o       |      |    Control Plane |       | Management Plane    |      |       | +----Y----+   +-----+   |       |  +-----+       +----Y----+ |       | | Service |   | App |   |       |  | App |       | Service | |       | +----Y----+   +--Y--+   |       |  +--Y--+       +----Y----+ |       |      |           |      |       |     |               |      |       | *----Y-----------Y----* |       | *---Y---------------Y----* |       | | Control Abstraction | |       | | Management Abstraction | |       | |     Layer (CAL)     | |       | |      Layer (MAL)       | |       | *----------Y----------* |       | *----------Y-------------* |       |            |            |       |            |               |       o------------|------------o       o------------|---------------o                    |                                 |                    | CP                              | MP                    | Southbound                      | Southbound                    | Interface                       | Interface                    |                                 |       *------------Y---------------------------------Y----------------*       |         Device and resource Abstraction Layer (DAL)           |       *------------Y---------------------------------Y----------------*       |            |                                 |                |       |    o-------Y----------o   +-----+   o--------Y----------o     |       |    | Forwarding Plane |   | App |   | Operational Plane |     |       |    o------------------o   +-----+   o-------------------o     |       |                       Network Device                          |       +---------------------------------------------------------------+
                                 Figure 2






Finn & Thubert         Expires September 10, 2015              [Page 10]Internet-Draft    Deterministic Networking Architecture       March 2015

4.1.  The Application Plane
   Per [RFC7426], the Application Plane includes both applications and   services.  In particular, the Application Plane incorporates the User   Agent, a specialized application that interacts with the end user /   operator and performs requests for Deterministic Networking services   via an abstract Stream Management Entity, (SME) which may or may not   be collocated with (one of) the end systems.
   At the Application Plane, a management interface enables the   negotiation of streams between end systems.  An abstraction of the   stream called a Traffic Specification (TSpec) provides the   representation.  This abstraction is used to place a reservation over   the (Northbound) Service Interface and within the Application plane.   It is associated with an abstraction of location, such as IP   addresses and DNS names, to identify the end systems and eventually   specify intermediate relay systems.
4.2.  The Controller Plane
   The Controller Plane corresponds to the aggregation of the Control   and Management Planes in [RFC7426], though Common Control and   Measurement Plane (CCAMP) [CCAMP] makes an additional distinction   between management and measurement.  When the logical separation of   the Control, Measurement and other Management entities is not   relevant, the term Controller Plane is used for simplicity to   represent them all, and the term controller refers to any device   operating in that plane, whether is it a Path Computation entity or a   Network Management entity (NME).  The Path Computation Element (PCE)   [PCE] is a core element of a controller, in charge of computing   Deterministic paths to be applied in the Network Plane.
   A (Northbound) Service Interface enables applications in the   Application Plane to communicate with the entities in the Controller   Plane.

TW> could you define NME, SME and PCE in the terminology?      One or more PCE(s) collaborate to implement the requests from the SME   as Per-Stream Per-Hop Behaviors installed in the relay systems for   each individual streams.  The PCEs place each stream along a   deterministic sequence of relay systems so as to respect per-stream   constraints such as security and latency, and optimize the overall   result for metrics such as an abstract aggregated cost.  The   deterministic sequence can typically be more complex than a direct   sequence and include redundancy path, with one or more packet   replication and elimination points.





Finn & Thubert         Expires September 10, 2015              [Page 11]Internet-Draft    Deterministic Networking Architecture       March 2015

4.3.  The Network Plane
   The Network Plane represents the network devices and protocols as a   whole, regardless of the Layer at which the network devices operate.
   The network Plane comprises the Network Interface Cards (NIC) in the   end systems, which are typically IP hosts, and relay systems, which   are typically IP routers and switches.  Network-to-Network Interfaces   such as used for Traffic Engineering path reservation in [RFC3209],   as well as User-to-Network Interfaces (UNI) such as provided by the   Local Management Interface (LMI) between network and end systems, are   all part of the Network Plane.
   A Southbound (Network) Interface enables the entities in the   Controller Plane to communicate with devices in the Network Plane.   This interface leverages and extends TEAS to describe the physical   topology and resources in the Network Plane.
                         Stream Management Entity
       End                                                     End           System                                               System
      -+-+-+-+-+-+-+ Northbound -+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
                PCE         PCE              PCE              PCE
      -+-+-+-+-+-+-+ Southbound -+-+-+-+-+-+-+-+-+-+--+-+-+-+-+-+-
                  Relay      Relay      Relay      Relay                  System     System     System     System       NIC                                                     NIC                  Relay      Relay      Relay      Relay                  System     System     System     System
                                 Figure 3
   The relay systems (and eventually the end systems NIC) expose their   capabilities and physical resources to the controller (the PCE), and   update the PCE with their dynamic perception of the topology, across   the Southbound Interface.  In return, the PCE(s) set the per-stream   paths up, providing a Stream Characterization that is more tightly   coupled to the relay system Operation than a TSpec.
   At the Network plane, relay systems exchange information regarding   the state of the paths, between adjacent systems and eventually with   the end systems, and forward packets within constraints associated to



Finn & Thubert         Expires September 10, 2015              [Page 12]Internet-Draft    Deterministic Networking Architecture       March 2015

   each stream, or, when unable to do so, perform a last resort   operation such as drop or declassify.
   This specification focuses on the Southbound interface and the   operation of the Network Plane.
4.4.  Elements of DetNet Architecture
   The DetNet architecture has a number of elements, discussed in the   following sections:
   a.  A model for the definition, identification, and operation of       DetNet streams (Section 4.5), for use by relay systems to       classify and process individual packets following per-stream       rules.TW> how is a packet "labeled" as part of stream?
   b.  A model for the flow of data from an end system or through a       relay system that can be used to predict the bounds for that       system's impact on the QoS of a DetNet stream, without       significantly constraining the method of implementing that       system, for use by the Controllers to configure policing and       shaping engines in Network Systems over the Southbound interface.       The model includes:
       1.  A model for queuing, transmission selection, shaping,           preemption, and timing resources that can be used by an end           system or relay system to control the selection of packets           output on an interface.  These models must have sufficiently           well-defined characteristics, both individually and in the           aggregate, to give predictable results for the QoS for DetNet           packets (Section 4.7).
       2.  A model for identifying misbehaving DetNet streams and           mitigating their impact on properly functioning streams           (Section 4.9).
   c.  A model for the relay system to inform the controller(s) of the       information it needs for adequate path computations including:
       1.  Systems' individual capabilities (e.g. can do replication,           can do precise time).
       2.  Link capabilities and resources (e.g. bandwidth, 0 delays,           hardware deterministic support to the physical layer, ...)
       3.  hysicalTW> typo           resources (total and available buffers, timers,           queues, etc)



Finn & Thubert         Expires September 10, 2015              [Page 13]Internet-Draft    Deterministic Networking Architecture       March 2015

       4.  Network Adjacencies (neighbors)
   d.  A model for the provision of a service, by end systems, or relay       systems, to forward a DetNet stream over a simple or redundant       path.  The model includes:
       1.  A model for an abstract relaying operation of either Routing           or forwarding packets of a DetNet stream to a next-hop relay           system, across Layer boundaries.
       2.  A model of next-hop(s) information for replicating the           packets of a DetNet stream, typically at or near the talker,           merging and/or re-replicating those packets at other points           in the network, and finally eliminating the duplicates,           typically at or near the listener(s), in order to provide           high availability (Section 3.3).
   e.  The protocol stack model for an end system and/or a relay system       should support the above elements in a manner that maximizes the       applicability of existing standards and protocols to the DetNet       problem, allows for the creation of new protocols where needed,       thus making DetNet an add-on feature to existing networks, rather       than a new way to do networking.  In particular this protocol       stack supports networks in which the path from talker to       listener(s) includes bridges and/or routers in any order       (Section 4.10).
   f.  A variety of models for the provisioning of DetNet streams can be       envisioned, including orchestration by a central controller or by       a federation of controllers, provisioning by relay systems and       end systems sharing peer-to-peer protocols, by off-line       configuration, or by a combination of these methods.  The       provisioning models are similar to existing Layer-2 and Layer-3       models, in order to minimize the amount of innovation required in       this area (Section 4.12).
4.5.  DetNet streams
4.5.1.  Talker guarantees
   DetNet streams can by synchronous or asynchronous.  The transmission   of packets in synchronous DetNet streams uses time synchronization   among the end and relay systems to control the flow of packets.   Asynchronous DetNet streams are characterized by:
   o  A maximum packet size;
   o  An observation interval; and


Finn & Thubert         Expires September 10, 2015              [Page 14]Internet-Draft    Deterministic Networking Architecture       March 2015

   o  A maximum number of transmissions during that observation      interval.
   These parameters, together with knowledge of the protocol stack used   (and thus the size of the various headers added to a packet), limit   the number of bit times per observation interval that the DetNet   stream can occupy the physical medium.
   The talker promises that these limits will not be exceeded.  If the   talker transmits less data than this limit allows, the unused   resources such as link bandwidth can be made available by the system   to non-DetNet packets.  However, making those resources available to   DetNet packets in other streams would serve no purpose.  Those other   streams have their own dedicated resources, on the assumption that   all DetNet streams can use all of their resources over a long period   of time.
   Note that there is no provision in DetNet for throttling streams; the   assumption is that a DetNet stream, to be useful, must be delivered   in its entirety.  That is, while any useful application is written to   expect a certain number of lost packets, the real-time applications   of interest to DetNet demand that the loss of data due to the network   is extraordinarily infrequent.
   Although DetNet strives to minimize the changes required of an   application to allow it to shift from a special-purpose digital   network to an Internet Protocol network, one fundamental shift in the   behavior of network applications that is impossible to avoid--the   reservation of resources before the application starts.  In the first   place, a network cannot deliver finite latency and practically zero   packet loss to an arbitrarily high offered load.  Secondly, achieving   practically zero packet loss for unthrottled (though bandwidth   limited) streams means that bridges and routers have to dedicate   buffer resources to specific streams or to classes of streams.  The   requirements of each reservation have to be translated into the   parameters that control each system's queuing, shaping, and   scheduling functions and delivered to the hosts, bridges, and   routers.
4.5.2.  Incomplete Networks
   The presence in the network of relay systems that are not fully   capable of offering DetNet services complicates the ability of the   relay systems and/or controller to allocate resources, as extra   buffering, and thus extra latency, must be allocated at each point   that is downstream from the non-DetNet relay system for some DetNet   stream.



Finn & Thubert         Expires September 10, 2015              [Page 15]Internet-Draft    Deterministic Networking Architecture       March 2015

4.6.  Data Flow Model through Systems
4.7.  Queuing, Shaping, Scheduling, and Preemption
   For this reason, the IEEE 802.1 Time-Sensitive Networking Task Group   has defined a set of queuing, shaping, and scheduling algorithms that   enable each bridge or router to compute the exact number of buffers   to be allocated for each stream or class of streams.
4.8.  Coexistence with normal traffic
   A DetNet network supports the dedication of at least 75% of the   network bandwidth to DetNet streams.  But, no matter how much is   dedicated for DetNet streams, It is zTW> typo   goal of DetNet to not interfere   excessively with existing QoS schemes.  It is also important that   non-DetNet traffic not disrupt the DetNet stream, of course (see   Section 4.9 and Section 6).  For these reasons:
   o  Bandwidth (transmission opportunities) not utilized by a DetNet      stream are available to non-DetNet packets (though not to other      DetNet streams).
   o  DetNet streams can be shaped, in order to ensure that the highest-      priority non-DetNet packet also is ensured a maximum latency.
   o  When transmission opportunities for DetNet streams are scheduled      in detail, then the algorithm constructing the schedule should      leave sufficient opportunities for non-DetNet packets to satisfy      the needs of the uses of the network.
   Ideally, the net effect of the presence of DetNet streams in a   network on the non-DetNet packets is primarily a reductoinTW> typo   in the   available bandwidth.
4.9.  Fault Mitigation
   One key to building robust real-time systems is to reduce the   infinite variety of possible failures to a number that can be   analyzed with reasonable confidence.  DetNet aids in the process by   providing filters and policers to detect DetNet packets received on   the wrong interface, or at the wrong time, or in too great a volume,   and to then take actions such as disabling the offending packet,   shutting down the offending DetNet stream, or shutting down the   offending interface.
   It is also essential that filters and service remarkingTW> please define "service remarking"   be employed   to prevent non-DetNet packets from impinging on the resources   allocated to DetNet packets.


Finn & Thubert         Expires September 10, 2015              [Page 16]Internet-Draft    Deterministic Networking Architecture       March 2015

   There exist techniques, at present and/or in various stages of   standardization, that can perform these fault mitigation tasks that   deliver a high probability that misbehaving systemdTW> typo   will have zero   impact on well-behaved DetNet streams, except of course, for the   receiving interface(s) immediately downstream of the misbehaving   device.
4.10.  Protocol Stack Model
   This section will be further developed.  See [IEEE802.1CB], Annex C,   for a description of the protocol stack.  This is very much a work in   progress, not a standard.  See also [IEEE802.1Qcc].
4.11.  Advertising resources, capabilities and adjacencies
4.12.  Provisioning model
4.12.1.  Centralized Path Computation and Installation
   A centralized routing model, such as provided with a PCE (RFC 4655   [RFC4655]), enables global and per-stream optimizations.  The model   is attractive but a number of issues are left to be solved.  In   particular:
   o  whether and how the path computation can be installed by 1) an end      device or 2) a Network Management entity,
   o  and how the path is set up, either by installing state at each hop      with a direct interaction between the forwarding device and the      PCE, or along a path by injecting a source-routed request at one      end of the path.
4.12.2.  Distributed Path Setup
   Whether a distributed alternative without a PCE can be valuable   should be studied as well.  Such an alternative could for instance   inherit from the Resource ReSerVation Protocol [RFC5127] (RSVP)   flows.
   In a Layer-2 only environment, or as part of a layered approach to a   mixed environment, IEEE 802.1 also has work, either completed or in   progress.  [IEEE802.1Q-2011] Clause 35 describes SRP, a peer-to-peer   protocol for Layer-2 roughly analogous to RSVP.  Almost complete is   [IEEE802.1Qca], which defines how ISIS can provide multiple disjoint   paths or distribution trees.  Also in progress is [IEEE802.1Qcc],   which expands the capabilities of SRP.




Finn & Thubert         Expires September 10, 2015              [Page 17]Internet-Draft    Deterministic Networking Architecture       March 2015

5.  Related IETF work
5.1.  Deterministic PHB
   [I-D.svshah-tsvwg-deterministic-forwarding] defines a Differentiated   Services Per-Hop-Behavior (PHB) Group called Deterministic Forwarding   (DF).  The document describes the purpose and semantics of this PHB.   It also describes creation and forwarding treatment of the service   class.  The document also describes how the code-point can be mapped   into one of the aggregated Diffserv service classes [RFC5127].
5.2.  6TiSCH
   Industrial process control already leverages deterministic wireless   Low power and Lossy Networks (LLNs) to interconnect critical   resource-constrained devices and form wireless mesh networks, with   standards such as [ISA100.11a] and [WirelessHART].
   These standards rely on variations of the [IEEE802154e] timeSlotted   Channel Hopping (TSCH) [I-D.ietf-6tisch-tsch] Medium Access Control   (MAC), and a form of centralized Path Computation Element (PCE), to   deliver deterministic capabilities.
   The TSCH MAC benefits include high reliability against interference,   low power consumption on characterized streams, and Traffic   Engineering capabilities.  Typical applications are open and closed   control loops, as well as supervisory control streams and management.
   The 6TiSCH Working Group focuses only on the TSCH mode of the IEEE   802.15.4e standard.  The WG currently defines a framework for   managing the TSCH schedule.  Future work will standardize   deterministic operations over so-called tracks as described in   [I-D.ietf-6tisch-architecture].  Tracks are an instance of a   deterministic path, and the DetNet work is a prerequisite to specify   track operations and serve process control applications.
   [RFC5673] and [I-D.ietf-roll-rpl-industrial-applicability] section   2.1.3.  and next discusses application-layer paradigms, such as   Source-sink (SS) that is a Multipeer to Multipeer (MP2MP) model that   is primarily used for alarms and alerts, Publish-subscribe (PS, or   pub/sub) that is typically used for sensor data, as well as Peer-to-   peer (P2P) and Peer-to-multipeer (P2MP) communications.  Additional   considerations on Duocast and its N-cast generalization are also   provided for improved reliability.






Finn & Thubert         Expires September 10, 2015              [Page 18]Internet-Draft    Deterministic Networking Architecture       March 2015

6.  Security Considerations
   Security in the context of Deterministic Networking has an added   dimension; the time of delivery of a packet can be just as important   as the contents of the packet, itself.  A man-in-the-middle attack,   for example, can impose, and then systematically adjust, additional   delays into a link, and thus disrupt or subvert a real-time   application without having to crack any encryption methods employed.   See [RFC7384] for an exploration of this issue in a related context.
   Furthermore, in a control system where millions of dollars of   equipment, or even human lives, can be lost if the DetNet QoS is not   delivered, one must consider not only simple equipment failures,   where the box or wire instantly becomes perfectly silent, but bizarre   errors such as can be cousedTW> typo   by software failures.  Because there is   essentiallTW> typo   no limit to the kinds of failures that can occur,   protecting against realistic equipment failures is indistinguishable,   in most cases, from protecting against malicious behavior, whether   accidental or intentional.  See also Section 4.9.
   Security must cover:
   o  the protection of the signaling protocol
   o  the authentication and authorization of the controlling systems
   o  the identification and shaping of the streams
7.  IANA Considerations
   This document does not require an action from IANA.
8.  Acknowledgements
   The authors wish to thank Jouni Korhonen, Erik Nordmark, George   Swallow, Rudy Klecka, Anca Zamfir, David Black, Thomas Watteyne,   Shitanshu Shah, Craig Gunther, Rodney Cummings, Wilfried Steiner,   Marcel Kiessling, Karl Weber, Ethan Grossman and Pat Thaler, for   their various contribution with this work.
9.  Informative References
   [AVnu]     http://www.avnu.org/, "The AVnu Alliance tests and              certifies devices for interoperability, providing a simple              and reliable networking solution for AV network              implementation based on the Audio Video Bridging (AVB)              standards.", .



Finn & Thubert         Expires September 10, 2015              [Page 19]Internet-Draft    Deterministic Networking Architecture       March 2015

   [CCAMP]    IETF, "Common Control and Measurement Plane",              <https://datatracker.ietf.org/doc/charter-ietf-ccamp/>.
   [HART]     www.hartcomm.org, "Highway Addressable Remote Transducer,              a group of specifications for industrial process and              control devices administered by the HART Foundation", .
   [HSR-PRP]  IEC, "High availability seamless redundancy (HSR) is a              further development of the PRP approach, although HSR              functions primarily as a protocol for creating media              redundancy while PRP, as described in the previous              section, creates network redundancy.  PRP and HSR are both              described in the IEC 62439 3 standard.",              <http://webstore.iec.ch/webstore/webstore.nsf/              artnum/046615!opendocument>.
   [I-D.finn-detnet-problem-statement]              Finn, N. and P. Thubert, "Deterministic Networking Problem              Statement", draft-finn-detnet-problem-statement-01 (work              in progress), October 2014.
   [I-D.ietf-6tisch-architecture]              Thubert, P., Watteyne, T., Struik, R., and M. Richardson,              "An Architecture for IPv6 over the TSCH mode of IEEE              802.15.4e", draft-ietf-6tisch-architecture-05 (work in              progress), January 2015.
   [I-D.ietf-6tisch-tsch]              Watteyne, T., Palattella, M., and L. Grieco, "Using              IEEE802.15.4e TSCH in an IoT context: Overview, Problem              Statement and Goals", draft-ietf-6tisch-tsch-05 (work in              progress), January 2015.
   [I-D.ietf-roll-rpl-industrial-applicability]              Phinney, T., Thubert, P., and R. Assimiti, "RPL              applicability in industrial networks", draft-ietf-roll-              rpl-industrial-applicability-02 (work in progress),              October 2013.
   [I-D.svshah-tsvwg-deterministic-forwarding]              Shah, S. and P. Thubert, "Deterministic Forwarding PHB",              draft-svshah-tsvwg-deterministic-forwarding-03 (work in              progress), March 2015.
   [IEEE802.1AS-2011]              IEEE, "Timing and Synchronizations (IEEE 802.1AS-2011)",              2011, <http://standards.ieee.org/getIEEE802/              download/802.1AS-2011.pdf>.


Finn & Thubert         Expires September 10, 2015              [Page 20]Internet-Draft    Deterministic Networking Architecture       March 2015

   [IEEE802.1BA-2011]              IEEE, "AVB Systems (IEEE 802.1BA-2011)", 2011,              <http://standards.ieee.org/getIEEE802/              download/802.1BA-2011.pdf>.
   [IEEE802.1CB]              IEEE, "Seamless Redundancy (IEEE Draft P802.1CB)", 2015,              <http://p8021:[e-mail]/1/files/private/              cb-drafts/>.
   [IEEE802.1Q-2011]              IEEE, "MAC Bridges and VLANs (IEEE 802.1Q-2011", 2011,              <http://standards.ieee.org/getIEEE802/              download/802.1Q-2011.pdf>.
   [IEEE802.1Qat-2010]              IEEE, "Stream Reservation Protocol (IEEE 802.1Qat-2010)",              2010, <http://standards.ieee.org/getIEEE802/              download/802.1Qat-2010.pdf>.
   [IEEE802.1Qav]              IEEE, "Forwarding and Queuing (IEEE 802.1Qav-2009)", 2009,              <http://standards.ieee.org/getIEEE802/              download/802.1Qav-2009.pdf>.
   [IEEE802.1Qca]              IEEE, "Path Control and Reservation", 2015,              <http://p8021:[e-mail]/1/files/private/              ca-drafts/>.
   [IEEE802.1Qcc]              IEEE, "Stream Reservation Protocol (SRP) Enhancements and              Performance Improvements", 2015,              <http://p8021:[e-mail]/1/files/private/              cc-drafts/>.
   [IEEE802.1TSNTG]              IEEE Standards Association, "IEEE 802.1 Time-Sensitive              Networks Task Group", 2013,              <http://www.IEEE802.org/1/pages/avbridges.html>.
   [IEEE802154]              IEEE standard for Information Technology, "IEEE std.              802.15.4, Part. 15.4: Wireless Medium Access Control (MAC)              and Physical Layer (PHY) Specifications for Low-Rate              Wireless Personal Area Networks", June 2011.




Finn & Thubert         Expires September 10, 2015              [Page 21]Internet-Draft    Deterministic Networking Architecture       March 2015

   [IEEE802154e]              IEEE standard for Information Technology, "IEEE std.              802.15.4e, Part. 15.4: Low-Rate Wireless Personal Area              Networks (LR-WPANs) Amendment 1: MAC sublayer", April              2012.
   [ISA100.11a]              ISA/IEC, "ISA100.11a, Wireless Systems for Automation,              also IEC 62734", 2011, < http://www.isa100wci.org/en-              US/Documents/PDF/3405-ISA100-WirelessSystems-Future-broch-              WEB-ETSI.aspx>.
   [ODVA]     http://www.odva.org/, "The organization that supports              network technologies built on the Common Industrial              Protocol (CIP) including EtherNet/IP.", .
   [PCE]      IETF, "Path Computation Element",              <https://datatracker.ietf.org/doc/charter-ietf-pce/>.
   [Profinet]              http://us.profinet.com/technology/profinet/, "PROFINET is              a standard for industrial networking in automation.",              <http://us.profinet.com/technology/profinet/>.
   [RFC2205]  Braden, B., Zhang, L., Berson, S., Herzog, S., and S.              Jamin, "Resource ReSerVation Protocol (RSVP) -- Version 1              Functional Specification", RFC 2205, September 1997.
   [RFC3209]  Awduche, D., Berger, L., Gan, D., Li, T., Srinivasan, V.,              and G. Swallow, "RSVP-TE: Extensions to RSVP for LSP              Tunnels", RFC 3209, December 2001.
   [RFC4655]  Farrel, A., Vasseur, J., and J. Ash, "A Path Computation              Element (PCE)-Based Architecture", RFC 4655, August 2006.
   [RFC5127]  Chan, K., Babiarz, J., and F. Baker, "Aggregation of              Diffserv Service Classes", RFC 5127, February 2008.
   [RFC5673]  Pister, K., Thubert, P., Dwars, S., and T. Phinney,              "Industrial Routing Requirements in Low-Power and Lossy              Networks", RFC 5673, October 2009.
   [RFC7384]  Mizrahi, T., "Security Requirements of Time Protocols in              Packet Switched Networks", RFC 7384, October 2014.






Finn & Thubert         Expires September 10, 2015              [Page 22]Internet-Draft    Deterministic Networking Architecture       March 2015

   [RFC7426]  Haleplidis, E., Pentikousis, K., Denazis, S., Hadi Salim,              J., Meyer, D., and O. Koufopavlou, "Software-Defined              Networking (SDN): Layers and Architecture Terminology",              RFC 7426, January 2015.
   [TEAS]     IETF, "Traffic Engineering Architecture and Signaling",              <https://datatracker.ietf.org/doc/charter-ietf-teas/>.
   [WirelessHART]              www.hartcomm.org, "Industrial Communication Networks -              Wireless Communication Network and Communication Profiles              - WirelessHART - IEC 62591", 2010.
Authors' Addresses
   Norm Finn   Cisco Systems   170 W Tasman Dr.   San Jose, California  95134   USA
   Phone: [phone]   Email: [e-mail]

   Pascal Thubert   Cisco Systems   Village d'Entreprises Green Side   400, Avenue de Roumanille   Batiment T3   Biot - Sophia Antipolis  06410   FRANCE
   Phone: [phone]   Email: [e-mail]















Finn & Thubert         Expires September 10, 2015              [Page 23]

_______________________________________________
6tisch mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/6tisch