Publish-Subscribe Deployment Option for NDN in the Constrained Internet of Things
draft-gundogan-icnrg-pub-iot-02

Abstract

Constrained IoT devices often operate more efficiently in a loosely coupled environment without maintaining end-to-end connectivity between nodes. Information Centric Networking naturally supports this demand by replicated data distribution and hop wise forwarding. This document outlines a deployment option for NDN in low-power and lossy networks (LLNs) that follows a publish-subscribe pattern. The proposed protocol scheme simplifies name-based routing significantly and facilitates even large off-duty cycles for constrained nodes.

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 https://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 6, 2018.

Copyright Notice

Copyright (c) 2018 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 (https://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.

Table of Contents

• 1. Introduction
• 1.1. Baseline Scenarios
• 1.2. Benefits of Loose Coupling in the IoT
• 2. Terminology
• 3. Publish-Subscribe in IoT Edge Networks
• 3.1. Topology Maintenance and Routing
• 3.2. Mapping Publish to NDN
• 3.3. Mapping Subscribe to NDN
• 3.4. Content Replication in partitioned networks
• 3.5. Content Replication between Proxy Instances
• 3.6. Alerting and group communication
• 4. Control Plane Messaging
• 4.1. Prefix Advertisement Message (PAM)
• 4.2. Name Advertisement Message (NAM)
• 4.2.1. Name Option
• 5. Security Considerations
• 6. References
• 6.1. Normative References
• 6.2. Informative References
• Acknowledgments
• Authors' Addresses

1. Introduction

In the emerging Internet of Things (IoT), it is expected that large quantities of very constrained sensors and actuators collect, communicate, and process massive amounts of machine data. Early experiments with constrained nodes show promising results for different deployments of ICN communication [NDN-EXP]. NDN stacks for the IoT tend to be less resource-consuming and more lightweight, the hop wise caching and forwarding indicate higher reliability and lower sensitivity to disruptions than end-to-end IP solutions. Most notably, security and resiliance of nodes and networks are enhanced by the NDN request-response scheme, where data delivery remains receiver-initiated.

The following properties are characteristic for constrained IoT environments:

• Battery-powered nodes with significant sleep cycles
• Low memory reserves, but potentially large (flash) storage
• Failing nodes due to various external or internal reasons
• Low power lossy wireless networks

These various constraints pose significant challenges on a real-world deployment of NDN networks, among them [RFC7927]:

• The complexity of name-based routing may overburden link and memory capacities.
• State management at nodes may drain batteries or memory resources.
• Clear separation between control and data plane potentially increases
• An adaptation to the constrained wireless link and transmission characteristics will often be required (see [I-D.gundogan-icnrg-ccnlowpan]).
• Mobility management will need to aid non-stationary deployments.
• Intermittent connectivity may occur from mobility or temporarily fragmenting networks.

1.1. Baseline Scenarios

Multiple scenarios have been discussed in [RFC7476] and [IWMT] that evaluate the applicability of ICN in IoT.

We consider two characteristic constrained IoT scenarios with the enumerated challenges below.

Stationary IoT nodes within reach of fixed uplinks

for home, building, and factory automation, stationary monitoring, and related stable environments:

• Reliability, resilience of operation
• Radio coordination, coverage
• Energy constraints, device lifetime
• Interference with rivaling appliances

Mobile IoT nodes with sparse coverage or intermittent connectivity

for urban or rural mobility and sensing, industrial Internet in widespread environments, disaster recovery and rescue, and related unstable environments:

• Exploit connectivity when available
• Large off-duty cycles of nodes
• Partitioned networks
• Limited dependability
• Environmental impact and disturbance

IoT scenarios usually impose routing requirements to support mobile nodes, handle failing links and to be resilient against attacks. A secure and autonomous bootstrapping is essential, especially for large-scale IoT deployments.

1.2. Benefits of Loose Coupling in the IoT

ICN decouples content consumers from data producers (decoupling in space). A more sophisticated decoupling can be provided with the publish-subscribe messaging pattern that further adds a decoupling in time and synchronization. Constrained devices in LLNs can leverage this loose coupling to increase sleep cycles and delegate the authority over as much information as possible to more powerful devices that act as content proxies. In Figure 1, once content is published to the content proxy (CP) by a producer (P), consumers (C) can retrieve this content from (CP) without interacting with the producer. This indirection when retrieving information allows (P) to align sleep cycles accordingly to the period of generating new sensor readings, instead of handling content requests from any consumers (C).

      (CP) 
     / | \ 
    /  |  `-----. 
   /   |  |  |   \ 
  (P) (C) ...... (C) 

Figure 1: Content Proxy (CP) - Producer (P) - Consumer (C)

The NDN communication follows a request-response pattern and prevents an unsolicited push of data. While this design must be seen as a vital security feature that prevents common DDoS attacks, no straight-forward mechanism for issuing an unsolicited alert between nodes has been foreseen which is desired in many use cases of sensors and actuators. Several extensions have been proposed to enable an unsolicited push of data, among them Interest-follows-Interest, Interest notification, and a dedicated push packet. All these push messages are sent immediately to a prospective consumer node, which not only conflicts with the ICN paradigm of naming content instead of hosts, but re-open the DoS attack surface.

In this memo, we introduce the different approach of a (local) 'Hop and Pull' [HoPP]. We assume the presence of a link-local control message for alerting next-hop neighbors. Such single-hop alerts advertise names (NAM-messages), or prefixes (PAM-mesage). Nodes that receive such local control information may either update their control plane (routing tables) or decide to pull content from its neighboring nodes. In this way, the NDN request-response pattern for data transmission remains fully intact, while control functions and time-critical update alerts become available as first class citizens in the ICN scheme.

2. Terminology

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 RFC 2119 [RFC2119]. The use of the term, "silently ignore" is not defined in RFC 2119. However, the term is used in this document and can be similarly construed.

This document uses the terminology of [RFC7476], [RFC7927], and [RFC7945] for ICN entities.

The following terms are used in the document and defined as follows:

Converge Cast

Many-to-One communication pattern, where multiple devices send sensor readings to one gateway.

Content Proxy

Stable node for replicating content.

Cloud Gateway

A Gateway that enables content transfer to and from a remote cloud storage, possibly by performing some kind of protocol translation.

PAM

Prefix Advertisement Message.

NAM

Name Advertisement Message.

3. Publish-Subscribe in IoT Edge Networks

The publish-subscribe system is centered around prefix-specific content proxies (CPs) that are deployed in IoT edge networks. Such proxy function can be hosted on the Cloud- or Internet Gateway, or may reside on a stable, less constrained node within the IoT infrastructure. It is assumed that a CP is present for each prefix covering publishable content.

Implementing a pub-sub NDN involves several steps that are bound to the tight requirements of resource-constrained devices. These steps include:

1. Building the prefix-specific routing topology tailored to constrained networks
2. Mapping Publish to NDN semantics
3. Mapping Subscribe to NDN semantics

3.1. Topology Maintenance and Routing

A (sensor) node that wants to publish a data item needs to rely on path information towards the Content Proxy. Following the approach of PANINI [PANINI], default routes will be established as follows.

Each CP in the local IoT sub-network advertises the prefix(es) it represents to the routing system. It does so by broadcasting Prefix Advertisement Messages (PAMs) on the link layer (see Section 4 for the corresponding protocol details). Nodes that newly receive PAM advertisements will add or refresh a prefix-specific default route in their FIB. Intermediate nodes in a multi-hop environment also re-broadcast PAMs, so that the entire sub-network is flooded and default route entries build a shortest path tree (SPT) towards the CP as shown in Table 1 (alternatively, a DODAG w.r.t. a gateway for redundant CPs).

                (CP)   
                /  \ 
          PAM  /    \ PAM 
             (A)    (B) 
             /|\    /|\ 
            : : :  : : : PAM 

Figure 2: SPT building by Prefix Advertisement Messages (PAMs)

Information flowing from constrained sensor nodes towards a gateway is the prevalent communication pattern in the IoT (converge cast). The publish-subscribe system hence establishes a default routing (see sample FIB in Table 1) and uses the tree topology with default routes towards the CP as a first step of content aggregation. Content replication towards other CPs, an Internet gateway, or into a cloud can follow subsequently.

It is noteworthy that the role of the new PAM message remains orthogonal to the existing Interest or Data semantics. A PAM never carries data nor requests, but resides on the control plane of name-based routing. User applications stay unaffected, and continue to rely on the NDN-specific request-response paradigm.

Each node in the tree calculates a rank based on the rank of its parent and a routing metric. Hence, the rank is an indication for the position within the tree and is strictly monotonically increasing in the downwards direction from root to leaf. For simplicity, this document uses the hop-count routing metric to calculate the rank. Future work can focus on more elaborate routing metrics, e.g., to reduce packet retransmissions or improve load balancing for publish operations.

The Routing Information Base (RIB) contains the following information:

SPT

Prefix:

Variable length prefix to configure a prefix-specific default route

Rank:

16-bit unsigned integer indicating a node's rank

Flags:

8-bit unsigned integer to store SPT related flags

Parent

The appropriate face towards the parent node is stored.

NAM Cache (NC)

Entries in the NAM Cache are <name,downstream_face> tuples. The NC is consolidated when propagating name advertisements.

3.2. Mapping Publish to NDN

In classical publish-subscribe systems, a Publish is typically implemented as a push mechanism on the data plane. However, this contradicts the request-response paradigm employed by NDN. To adapt the Publish operation to NDN semantics, it is split into two phases and the required push mechanism is performed on the control plane with a link-local scope. The two phases consist of:

1. Announcing names of Named Data Objects (NDO) to neighbors on the control plane
2. Requesting NDOs on the data plane

In the first phase, the name of the NDO to publish is advertised to the prefix-specific upstream neighbor by encoding it with TLV format in a unicast link-local Name Advertisement Message (NAM) adopted from PANINI [PANINI]. Once an upstream neighbor receives an unsolicited NAM, the name is extracted and along with the incoming face stored in the NAM Cache (NC) as a <name,downstream_face> tuple. This link-local content signaling does not establish any data paths in the PIT, nor does it modify the FIB or the Content Store (CS).

In the second phase and as a result of an unsolicited NAM, the content is requested from the downstream neighbor according to the standard NDN scheme with an Interest message for the name recorded in the NC on the recorded face (see [HoPP]). When a downstream neighbor replies with the content (Data message), this neighbor removes the corresponding NC entry and the NDO to publish is successfully replicated one hop closer towards the content proxy. NC entries are thus short-lived for hop-wise replication only.

Both phases are iteratively repeated hop-wise until the content proxy is reached. Being the root of this prefix-specific spanning tree, the content proxy has no further prefix-specific upstream neighbor and the replication terminates. It is noteworthy, that both phases can be interleaved, so that NAMs are signaled to the upstream neighbor, while the content is requested from the downstream neighbor.

Figure 3 (a) depicts the hop-wise replication of a published content on the gradient towards the (CP). In this example, the name /HAW/temp_t is advertised by the producer (P) to its parent node (A). (A) extracts the name from the NAM and stores it along with the incoming face in its NC. (A) then propagates the NAM further to its parent (CP) and simultaneously requests the content from (P) as depicted in Figure 3 (b). Likewise in Figure 3 (c), (CP) reacts to the incoming NAM by requesting the content from (A). NC states from (A) and (P) are removed as soon as they respond to the request.

 +-------------+ +------------------------+ +-------------------+ 
 |             | |                        | | Interest          | 
 |    (CP)     | |            ,->(CP)     | |    ,----(CP)<-,   | 
 |    /        | |        NAM |  /        | |    |   /     /    | 
 |   /         | |            | /         | |    |  /     /     | 
 | (A)<-,      | |          ,-(A)<-,      | |    `>(A)---' Data | 
 |  \   | NAM  | |          |  \   | Data | |        \          | 
 |   \  |      | |          |   \  |      | |         \         | 
 |    (P)      | | Interest `--->(P)      | |         (P)       | 
 | /HAW/temp_t | |            /HAW/temp_t | |      /HAW/temp_t  | 
 +-------------+ +------------------------+ +-------------------+ 
  (a) Adv. Name    (b) Replicate Content     (c) Finish publish 

Figure 3: Publish

3.3. Mapping Subscribe to NDN

In the proposed publish-subscribe system, the Subscribe operation is equivalent to an Interest-based request of previously learned content names. A device can learn about new content in various ways:

1. Name Advertisements by the CP via a dedicated prefix path or broadcast.
2. By requesting a named topic from the CP and receiving a specific name of available content in return.
3. By polling a name inventory / a dedicated data structure from the CP that records publishings and updates. Such a data structure may contain indicators about the periodicity of sensor readings to align periodic polling schemes to sensor reading intervals.
4. By (offline) encoding topics into a hierarchical naming scheme of the form routing prefix / topic / unique_part.

Consumers request topics as well as named content by regurlar Interest requests. Thereby, Inerest timeouts for these requests serve as subscription periods, may be selected of a specific (long) timespan, and must be refreshed or retriggered for every publishing to adhere to the flow control approach of NDN.

3.4. Content Replication in partitioned networks

The hop-wise replication described in Section 3.2 transparently supports publish operations in partitioned networks. When a publish operation fails to replicate content and no backup parent is in the vicinity (Figure 4 (b)), the node marks its sub-DAG as floating by propagating PAMs with the floating bit set and the NC entry is preserved. Once a sub-DAG becomes connected to another parent, the publishing is resumed (Figure 4 (c)).

 +-------------+     +-------------+    +-------------+ 
 |             |     |             |    |             | 
 |    (CP)     |     |     (CP)    |    |    (CP)<-,  | 
 |             |     |             |    |    /    /   | 
 |             |     |         X   |    |   /    /    | 
 |  (A)<-,     |     |  (A)---'    |    | (A)---'     | 
 |   \   | PUB |     |   \    PUB  |    |  \    PUB   | 
 |    \  |     |     |    \        |    |   \         | 
 |     (P)     |     |    (P)      |    |   (P)       | 
 | /HAW/temp_t |     | /HAW/temp_t |    | /HAW/temp_t | 
 +-------------+     +-------------+    +-------------+ 
(a) Publish content (b) Publish fails (c) Finish Publish 

Figure 4: Publish in partitioned network

A node receiving a PAM from its parent with the floating bit set may rejoin the tree using another parent in the neighborhood that is connected to the content proxy. Rejoining the tree may result in a worse rank.

3.5. Content Replication between Proxy Instances

Content Proxies within a network domain MAY advertise prefixes from an overlapping prefix range and thereby implement multiple upstreams in (some of) the FIBs. According to Section 3.2, content will be published to these multiple CPs upstream and thus populate a replicated environment.

Content Proxies MAY equally establish a subscription hierarchy between neighboring domains. This requires an exchange of corresponding FIB entries between the involved domain. It is noteworthy that a global exchange of FIB tables (e.g., Inter-provider peering) is beyond the scope of this document.

3.6. Alerting and group communication

TODO

4. Control Plane Messaging

Control plane messaging is currently under separate discussion. Details will be added after consensus is found.

TODO: add mappings of PAM and NAM to the NDN and CCNx dialects

4.1. Prefix Advertisement Message (PAM)

  0                   1                   2                   3 
  0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 
 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 
 | Type = PAM    |F|   Reserved  |             Rank              | 
 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 
 |        Prefix Length          |            Prefix ...          
 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 

Figure 5: Prefix Advertisement Message (PAM)

Message type:

8-bit unsigned integer. Indicates a PAM.

Floating (F):

1-bit floating flag. Indicates that a sub-tree is not connected to the content proxy, when set.

Reserved:

7-bit unused field. It MUST be initialized to zero by the sender and MUST be ignored by the receiver.

Rank:

16-bit unsigned integer. Indicates a node's position in the SPT.

Prefix Length:

16-bit unsigned integer. Indicates the length of the following prefix in bytes.

Prefix:

Variable length prefix used to configure a default route within the SPT.

4.2. Name Advertisement Message (NAM)

 0                   1                   2                   3 
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 
|  Type = NAM   |   Reserved    |        Options Length         | 
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 
:                            Options                            : 
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 

Figure 6: Name Advertisement Message (NAM)

Type:

8-bit unsigned integer. Indicates a NAM.

Reserved:

8-bit unused field. It MUST be initialized to zero by the sender and MUST be ignored by the receiver.

Length:

16-bit unsigned integer. Indicates the accumulated length of all options.

4.2.1. Name Option

 0                   1                   2                   3  
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 
|  Type = 0x00  |          Name Length          |   Name ... 
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 

Figure 7: Name option format

Type:

8-bit unsigned integer. Indicates the name option (0x00).

Name Length:

16-bit unsigned integer. Indicates the length of the name, excluding the type and length field.

Name:

Variable length name.

5. Security Considerations

TODO

6. References

6.1. Normative References

6.2. Informative References

Acknowledgments

This work was stimulated by fruitful discussions in the ICNRG, several research project teams and in personal communication. We would like to thank all active members for constructive thoughts and feedback. In particular, the authors would like to thank (in alphabetical order) Emmanuel Baccelli, Michael Frey, Oliver Hahm, Peter Kietzmann, Dirk Kutscher, Martine Lenders, Dave Oran, Joerg Ott, Hauke Petersen, Felix Shzu-Juraschek, and Christian Tschudin. This work was partly funded by the German Federal Ministry of Education and Research, project I3.

Authors' Addresses