Orchestra: Robust Mesh NetworksThrough Autonomously Scheduled TSCH

Simon Duquennoy1simonduq@sics.se

Beshr Al Nahas2beshr@chalmers.se

Olaf Landsiedel2olafl@chalmers.se

Thomas Watteyne3thomas.watteyne@inria.fr

1SICS Swedish ICT, Sweden2Chalmers University of Technology, Sweden

3Inria, France

ABSTRACTTime slotted operation is a well-proven approach to achievehighly-reliable low-power networking through schedulingand channel hopping. It is, however, difficult to apply timeslotting to dynamic networks as envisioned in the Internetof Things. Commonly, these applications do not have pre-defined periodic traffic patterns and nodes can be added orremoved dynamically.

This paper addresses the challenge of bringing TSCH(Time Slotted Channel Hopping MAC) to such dynamicnetworks. We focus on low-power IPv6 and RPL networks,and introduce Orchestra. In Orchestra, nodes autonomouslycompute their own, local schedules. They maintain multipleschedules, each allocated to a particular traffic plane (ap-plication, routing, MAC), and updated automatically as thetopology evolves. Orchestra (re)computes local scheduleswithout signaling overhead, and does not require any cen-tral or distributed scheduler. Instead, it relies on the existingnetwork stack information to maintain the schedules. Thisscheme allows Orchestra to build non-deterministic networkswhile exploiting the robustness of TSCH.

We demonstrate the practicality of Orchestra and quantifyits benefits through extensive evaluation in two testbeds, ontwo hardware platforms. Orchestra reduces, or even elim-inates, network contention. In long running experimentsof up to 72 h we show that Orchestra achieves end-to-enddelivery ratios of over 99.99%. Compared to RPL in asyn-chronous low-power listening networks, Orchestra improvesreliability by two orders of magnitude, while achieving asimilar latency-energy balance.

Categories and Subject DescriptorsC.2.1 [Network Architecture and Design]: WirelessCommunication

KeywordsTSCH, RPL, Scheduling, Wireless Sensor Network

Permission to make digital or hard copies of all or part of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies bear this notice and the full citationon the first page. Copyrights for components of this work owned by others than theauthor(s) must be honored. Abstracting with credit is permitted. To copy otherwise, orrepublish, to post on servers or to redistribute to lists, requires prior specific permissionand/or a fee. Request permissions from Permissions@acm.org.SenSys15, November 14, 2015, Seoul, South Korea..Copyright is held by the owner/author(s). Publication rights licensed to ACM.ACM 978-1-4503-3631-4/15/11 ...$15.00.DOI: http://dx.doi.org/10.1145/2809695.2809714.

TSCH beacon schedule RPL traffic schedule

Application traffic schedule

Orchestra schedule

1 slot = 10ms

Time

Figure 1: In Orchestra, several schedules repeatat different periods (here 2, 3, and 5), with slotsallocated to specific traffic planes (resp. Applica-tion, RPL, TSCH). Slots are skipped whenever over-lapped by a slot in a higher priority schedule (pri-ority increasing from top to bottom in the figure).

1. INTRODUCTIONContext. As the Internet of Things (IoT) is emerging,there is an increasing need for low-power communicationsolutions that are both flexible (i.e., are easy to use andable to to satisfy a variety of often dynamic application re-quirements) and robust (i.e., work reliably). Example ap-plications range from smart homes to smart cities, includ-ing wearable consumer devices. In these scenarios, short-range, low-power mesh networking is envisioned as a candi-date technology to achieve both energy-efficiency and reli-able large-scale operation.Challenge. Flexibility and reliability are opposing goals.Asynchronous low-power mesh networks (including low-power IPv6) are flexible and support non-deterministic ap-plications, but are best-effort. State-of-the art solutionshave loss rates in the range of one percent [15, 19, 21, 11].In the absence of end-to-end reliability, i.e., transport layerre-transmissions, such a loss rate is too high for most ap-plications. With end-to-end reliability, losses trigger costlyre-transmissions which often come in burst and result injittery performance. At the other end of the spectrum, de-terministic networks running TDMA and scheduled trafficcan achieve 2 or 3 orders of magnitude fewer losses (i.e., upto one loss per 10.000 packets or more) [8, 2, 37, 12, 28].We investigate how to achieve such high level of reliabilityin non-deterministic scenarios.Approach and Distinction. In this paper, we makea case for autonomous TSCH (Time Slotted Channel Hop-ping [1]) scheduling in non-deterministic low-power RPL andIPv6 networks. We show that even though it requires globalsynchronization, TSCH is practical in sparse traffic scenar-ios, and helps achieve high reliability in networks running adistributed routing protocol such as RPL [40]. The key chal-

lenge we address is that of creating TSCH schedules withouthindering any of the flexibility of RPL networks and match-ing the requirements of non-deterministic applications. Or-chestra achieves this with local, autonomous scheduling, andrequires neither a centralized nor a distributed scheduler.

This is radically different from traditional TSCHscheduling solutions such as WirelessHART [14] andISA100.11a [17], which rely on a centralized scheduling en-tity. This is also different from the standards being devel-oped in the IETF 6TiSCH working group [33], which employschedule negotiation between neighbor nodes.

In Orchestra, nodes employ simple periodic schedules andupdate the schedules automatically and instantly as therouting topology evolves. A TSCH schedule in Orchestraconsists of a set of over-provisioned communication slotsdedicated each to a specific communication plane: MAC,routing and application (as illustrated in Figure 1). As a re-sult, Orchestra allows building a generic, flexible, low-powerrouting backbone using RPL while benefiting from the ro-bustness of TSCH. Our schedules allow to reduce contentiondrastically, or even eliminate it altogether in certain cases.Results. We implement Orchestra and TSCH in Con-tiki [10], and experiment in two testbeds with 98 and25 nodes, each with a different hardware platform. In total,our evaluation bases on 219 individual experiments, up to72 hours long, and a total of 1,178,601 UDP packets routedfrom source to destination. We show that Orchestra en-ables autonomous TSCH scheduling in RPL networks, andachieves end-to-end delivery ratios over 99.99%. This is animprovement of 1 or 2 orders of magnitude over state-of-the-art asynchronous solutions such as RPL with ContikiMAC.We show that Orchestra achieves this strong reliability whilekeeping energy and latency close to the state of the art.Contribution. The main contribution of this paper is Or-chestra, a system that allows TSCH nodes to maintain theirschedules autonomously, driven by the state of the routingprotocol. Orchestra operates without a centralized sched-uler, and without inter-node schedule negotiation nor pathreservation. We demonstrate experimentally that Orchestrais practical, scalable, and achieves end-to-end loss rates twoorders of magnitude below low-power listening.Outline. The remainder of this paper is organized as fol-lows. 2 gives necessary background on TSCH and RPL,before introducing the basic concepts of Orchestra. 3 char-acterizes the potential benefits of TSCH as an alternativeto asynchronous MAC layers. 4 discusses the design of Or-chestra and 5 details implementation aspects. 6 discussesthe results of our thorough experimental evaluation in twodifferent testbeds as well as in controlled simulation. Wediscuss related work in 7 and conclude in 8.

2. OVERVIEWThis section introduces necessary background and gives a

brief overview of our system, Orchestra.

2.1 TSCHThe IEEE802.15.4e-2012 [1] standard defines a number of

MAC protocols for IEEE802.15.4. In this paper, we focuson TSCH (Time Slotted Channel Hopping), which inheritsfrom WirelessHART and ISA100.11a.

TSCH nodes form a globally synchronized low-power meshnetwork. Nodes may join the network after hearing an En-hanced Beacon (EB) from another node. Time synchroniza-

tion trickles from the PAN coordinator down to leaf nodesalong a Directed Acyclic Graph (DAG) structure. Time iscut into timeslots; timeslots are grouped into one or moreslotframes. A timeslot, typically 10 ms long, is long enoughfor a node to send a frame and for the receiver to acknowl-edge it. A TSCH schedule indicates to a node what to doin each timeslot: transmit, receive or sleep. A timeslot ina slotframe is identified by its time offset (when in the slot-frame it occurs), its channel offset (denoting the frequencyto communicate on), and a set of properties: whether it isto be used for transmission, reception, time synchronization,etc. Slots can be dedicated or shared, i.e., contention-freeor contention-based with CSMA back-off.

TSCH networks use channel hopping: the same slot inthe schedule translates into a different frequency at each it-eration of the slotframe. The result is that successive pack-ets exchanged between neighbor nodes are communicated atdifferent frequencies. In case a transmission fails because ofexternal interference or multi-path fading, its retransmissionhappens on a different frequency, often with a better proba-bility of succeeding than using the same frequency again [37].

How the communication schedule in the TSCH network isbuilt and maintained is out of the scope of the establishedstandards. The traditional way to scheduling (used in Wire-lessHART, ISA100.11a, and one of the modes in 6TiSCH) is