RTO considerations in LPWAN

Low-Power Wide Area Network (LPWAN) technologies offer appealing features, such as multikilometer wireless link range, while allowing low energy consumption for Internet of Things (IoT) devices. However, these advantages come at the expense of reduced physical layer (PHY) bit and message rates, which in some regions are further affected by spectrum access regulatory constraints. In some LPWAN scenarios, with flagship LPWAN technologies such as LoRaWAN or Sigfox, PHY bit rates are lower than 1 kbit/s, and uplink message rates are lower than 1 message/minute . Due to the aforementioned communication constraints, LPWAN technologies often exhibit high or very high Round Trip Times (RTTs). Even with negligible processing delays and in absence of communication errors, RTTs can be in the order of a few seconds or a few tens of seconds. Depending on the approach used to comply with spectrum access regulations, RTTs can grow to several minutes. Finally, when downlink responses are buffered in the radio gateway, RTTs will be in the order of the time between uplink messages (e.g. hours, if that is the time between two consecutive uplink messages). The described RTTs, as well as their potential variability, are significantly greater than typical ones on the Internet. In TCP, the default RTO used to be 3 seconds and was reduced to 1 second . In a similar order, the Constrained Application Protocol (CoAP), which is the preferred application-layer protocol for IPv6-based LPWAN, has a default RTO randomly chosen between 2 and 3 seconds . At the adaptation layer between IPv6 and the LPWAN technology, some of the Static Context Header Compression (SCHC) fragmentation modes also use RTOs, which need to be defined suitably for each LPWAN technology . This document provides guidance for suitable RTO configuration and/or usage in LPWAN. First, the document characterizes the RTT for LoRaWAN and Sigfox in absence of communication errors, buffering delays or processing delays. Second, higher order RTTs are described, capturing the impact of message rate limitations due to regulatory constraints and radio gateway buffering delays. Finally, the document discusses suitable RTO settings in LPWAN, and describes an experimental LPWAN-specific dual RTO algorithm.

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 .

This section provides an analysis of the RTT for relevant LPWAN technologies, such as LoRaWAN and Sigfox, assuming ideal conditions (i.e. no losses, as well as negligible buffering and processing delay). For detailed descriptions of LoRaWAN and Sigfox, the reader may refer to the literature . In the analysis, the RTT comprises the time since the start of the transmission of an uplink message by an IoT device until a response is completely received by the IoT device. A 4-byte SCHC-compressed IPv6/UDP/CoAP packet is assumed for the downlink response. Of course, larger sized packets will lead to greater RTTs.

shows the minimum and maximum theoretical RTT values for LoRaWAN in the EU band in ideal conditions. For the minimum ones, we assume a 4-byte uplink frame payload, and a downlink response sent in the first receive window. For the maximum ones, we assume the maximum allowed uplink payload size for each Data Rate (DR), and a downlink response sent in the second receive window. Note that there is a 1- or 2-second delay between the uplink transmission and the first or second receive window, respectively.

As shown in , and under the conditions assumed, the minimum RTT value for DR0 will always (for DR1, will almost always) exceed the default CoAP RTO. The maximum RTT will always exceed the default CoAP RTO for DR0-DR2, and will often exceed the default CoAP RTO for DR3-DR7. Note that since DR6 and DR7 are optional, they are not necessarily supported in real deployments.

shows the minimum and maximum theoretical RTT values for Sigfox in ideal conditions. For the minimum ones, we assume a 4-byte uplink frame payload, and a downlink response sent right at the beginning of the downlink receive window. For the maximum ones, we assume the maximum allowed uplink payload size, and a downlink response sent at the end of the receive window. Note that there is a 20-second delay between the frame uplink transmission and the start of the downlink receive window.

As shown in , and under the conditions assumed, the RTT in Sigfox will be one order of magnitude greater than the default CoAP RTO for all uplink bit rates and uplink frame payload sizes.

The high RTTs found in ideal conditions can be further exacerbated by two further behaviours of LPWAN networks: i) policies for compliance with duty cycle constraints, and ii) radio gateway buffering delays. EU spectrum access regulations for some ISM bands used by LPWAN technologies state that, unless listen-before-talk is used, the duty cycle needs to be lower than some limit (e.g. 1% in some frequency bands). Both LoRaWAN and Sigfox need to comply with such regulations. There may be different applicable policies intended to ensure compliance with the regulations. In one of them, in order to comply with the 1% duty cycle limitation, after sending an uplink frame, an IoT device keeps an idle period equal to 99 times the transmission time of the uplink frame. Such a policy may increase the RTT by up to two orders of magnitude. For example, in LoRaWAN, this policy leads to RTTs that will always exceed the default CoAP RTO, leading to an RTT of up to 282 seconds in the worst case. Another phenomenon that may happen in LPWAN relates with the fact that in some technologies and scenarios (e.g. the most typical LoRaWAN class, called class A, and in Sigfox), a downlink frame can only be sent during a given time interval (called receive window) after the uplink frame transmission. If a radio gateway misses the opportunity to send a downlink response to an uplink frame (e.g. because the radio gateway is busy sending other downlink messages or because it needs to refrain from transmitting immediately in order to comply with duty cycle regulations), the response to an uplink frame may be queued by the radio gateway until the next opportunity for sending a downlink frame. This problem has already been described in . If the problem occurs, the RTT will be tied to the time between two uplink consecutive frames. Depending on the application and its traffic pattern, such time may take values in the order of seconds, minutes, hours or even days.

The RTO needs to be greater than the RTT in order to avoid spurious timeouts. The latter are particularly expensive in LPWAN due to the message rate constraints in these networks, and also since they consume energy unnecessarily. However, as stated in , "each implementation of a retransmission timeout mechanism represents a balance between correctness and timeliness and therefore no implementation suits all situations". If delay is not relevant for an application, setting the default RTO to at least the highest frequently expected RTT, denoted HIGH_RTT, may be a suitable approach. The problem arises when delay, even if at LPWAN scales, matters, and higher order RTTs (Section 3) are expected in addition to the ideal scenario ones (Section 2). At the very least, the default RTO needs to be greater than the corresponding ideal scenario RTT value shown in Section 2. If higher order RTTs are expected, one option is using a simple dual RTO approach as follows. The LPWAN device keeps two RTO instances. One instance (called Low RTO) is initialized to a suitable ideal scenario RTT, denoted LOW_RTT. The other instance (called High RTO) is initialized to a value of at least HIGH_RTT. The dual RTO operates as follows (see ): Initially, the LPWAN device uses the High RTO. When the device uses the High RTO, after N_THRESH_LOW consecutive RTT samples lower than THRESH_LOW_RTT, the device switches to using the Low RTO. When the device uses the Low RTO, after N_THRESH_HIGH consecutive RTT samples greater than THRESH_HIGH_RTT, the device switches back to using the High RTO.

| High RTO |----+ | +----------+ | if N_THRESH_HIGH | | if N_THRESH_LOW consecutive RTT samples | | consecutive RTT samples > THRESH_HIGH_RTT | | < THRESH_LOW_RTT | +----------+ | +----| Low RTO |<---+ +----------+ ]]> The above described dual RTO algorithm may be applied to different RTO approaches, such as a constant RTO, a constant but dithered RTO (e.g. as in default CoAP), an adaptive RTO algorithm (e.g. as in TCP or CoCoA ), etc. If an adaptive RTO is used, performance will benefit from separating lower RTT and higher RTT regimes, avoiding inaccuracy due to a too high RTT variance. Note that the phenomena described in Section 3 are expected to yield systematically large step function RTT distributions. These deviate significantly from the roughly normal/gaussian RTT statistics assumed by the TCP RTO algorithm. Further refinement of the mechanism, to be discussed.

TBD

Carles Gomez has been funded in part by the Spanish Government (Ministerio de Ciencia, Innovacion y Universidades) through the Jose Castillejo grant CAS18/00170 and by European Regional Development Fund (ERDF) and the Spanish Government through project TEC2016-79988-P, AEI/FEDER, UE. His contribution to this work has been carried out during his stay as a visiting scholar at the Computer Laboratory of the University of Cambridge.