| ICN Research Group | C. Gundogan |
| Internet-Draft | TC. Schmidt |
| Intended status: Experimental | HAW Hamburg |
| Expires: September 29, 2019 | M. Waehlisch |
| link-lab & FU Berlin | |
| M. Frey | |
| F. Shzu-Juraschek | |
| Safety IO | |
| J. Pfender | |
| VUW | |
| March 28, 2019 |
Quality of Service for ICN in the IoT
draft-gundogan-icnrg-iotqos-00
This document describes manageable resources in ICN IoT deployments and a lightweight traffic classification method for mapping priorities to resources. Management methods are further derived for controlling latency and reliability of traffic flows in constrained environments.
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 29, 2019.
Copyright (c) 2019 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.
The performance of networked systems is largely determined by the resources available for forwarding messages between components. In addition to link capacities and buffer queues, Information-centric Networks rely on additional resources that shape its overall performance, namely Pending Interest Table space, and caching capacity.
Typical IoT deployments add tight resource constraints to this picture [RFC7228]: Nodes have processing and memory limitations, the underlying link layer technologies are lossy and restricted in bandwidth. Particularly in multi-hop networks, such constraints affect the overall performance, create bottlenecks, but may lead to cascading packet loss or energy depletion when PIT resources are independently evicted and forwarding states decorrelate [DECORRELATION]. Overprovisioning to counter performance flaws is infeasible for many IoT scenarios as it inflicts with use cases and increases deployment costs. Quality of Service (QoS) is a method to enhance overall performance by redistributing resources to a subset of messages, and - in the constrained IoT use case - to coordinate operations under resource scarcity.
IoT applications follow various use cases, of which different QoS requirements can be derived. While periodic sensor readings often comply with unmanaged performance, industrial control messaging or security alerts require (very) low latency, and safety-critical environmental recording or network management need (highly) reliable network services. Both quality levels can only be assured by appropriate resource reservations.
In order to achieve a QoS-aware information-centric IoT deployment, this document describes manageable resources in Section 3, defines a flow classification method in Section 4, and specifies priorities and their mappings in Section 5.
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:
The following resources contribute to the overall network performance in Information-Centric IoT Networking and need management for QoS control.
The link layer manages access to the media and provides space to buffer packets. Low latency applications require that requests are prioritized compared to regular priority data. Based on the request response pattern of ICN, link layer resources can be preallocated for data packets.
The Pending Interest Table (PIT) stores open requests at each hop. PIT resources are allocated when requests are forwarded, and they are released on returning responses.
Placement and replacement strategies of PIT entries directly influence the latency and reliability properties of traffic flows and thus should consider prioritization schemes. If the PIT is not saturated new PIT entries can be added. If the PIT is saturated, requests with higher priority should replace requests with lower priority to prevent higher latencies due to retransmissions.
Content stores (CS) enable transparent in-network caching and thus improve the transport in wireless and lossy environments by reducing hop traversals for content requests [NDN-EXP].
Placement and replacement strategies of data in content stores can affect the latency and reliability properties of traffic flows. The latency can be reduced by placing data closer to the consumers. Reliability can be improved by replicating data in multiple content stores to be resilient to node failures.
This document defines a traffic flow classification mechanism that aggregates names into equivalence classes in order to apply resource allocation decisions on messages of particular traffic flows.
A traffic class is a name prefix and each device maintains a list of valid classes. The actual distribution of traffic classes is out of scope of this document. The classes for request and response messages are derived by performing a longest prefix match (LPM) with the list of valid traffic classes and the Name TLV of the message. Examples are given in Figure 1.
list =
["/org", "/org /Hamburg", "/org /Berlin", "/org /Berlin /sensor" ]
LPM("/com" ,list) = ""
LPM("/org /Germany" ,list) = "/org"
LPM("/org /Hamburg" ,list) = "/org /Hamburg"
LPM("/org /Berlin /sensor /temp",list) = "/org /Berlin /sensor"
Figure 1: Example traffic flow class matches.
The empty traffic class "" is the default class for messages that do not match any valid traffic classes in the class list.
We define two priority levels to set the priorities for traffic flows in regards to latency and reliability.
Each list entry of the traffic class list from Section 4 has an associated priority tuple which distinctly specifies priority levels for the resources in Section 3. The tuple is of the following form:
priority tuple = < LATENCY_PRIORITY, RELIABILITY_PRIORITY >
Figure 2: Schema of the priority tuple.
As described above, the link layer provides space to buffer outgoing packets. For the two latency priorities, this space can be allocated into the following two queues:
Packets will be appended to the queue corresponding to their priority level.
In unsatured PITs, requests are added as new entries regardless of the priority level. In saturated PITs, EXPEDITED traffic replaces PIT entries of REGULAR traffic. If all entries in a saturated PIT are of a higher priority than the incoming request, then we RECOMMEND to drop the incoming request. If a saturated PIT contains entries of the same priority as the incoming request, we RECOMMEND to drop the incoming request to avoid cancelling active but incomplete ICN operations.
Nodes MAY implement a caching decision strategy instead of always caching all incoming content objects [ICN-CACHING]. If they do, the caching decision strategy MUST take the content object priority into account, such that lower priority content is not cached if the cache is saturated with higher priority content.
In unsaturated content stores, all content objects are passed to the cache decision strategy.
In saturated content stores, reliable traffic flows MUST be passed to the cache replacement strategy. Content objects with regular reliability requirements MUST be evicted first to make room for higher reliability content objects. Traffic flows with regular latency and regular reliability requirements MUST be passed to the cache replacement strategy. The cache replacement strategy MUST NOT evict content objects of higher reliability. Expedited traffic flows with regular reliability MUST be passed to the cache replacement strategy. Content objects with regular latency and regular reliability requirements MUST be evicted first, if an open PIT state is available. Otherwise, if no PIT state is available, then the cache replacement strategy MAY replace content objects of expedited or regular latency requirements and with regular reliability requirements.
TODO
TODO
This work was stimulated by fruitful discussions in the ICNRG research group. We would like to thank all active members for constructive thoughts and feedback. In particular, the authors would like to thank Ilya Moiseenko and Dave Oran for their work provided in [I-D.moiseenko-icnrg-flowclass]. This work was supported in part by the German Federal Ministry of Research and Education within the I3 project.