1089-7798 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LCOMM.2017.2764889, IEEE Communications Letters 1 MDP-based Model for Interest Scheduling in IoT-NDN Environment Shapna Muralidharan, Abhishek Roy and Navrati Saxena Abstract—Named Data Networking (NDN) is a novel network- ing paradigm that can acknowledge the unprecedented increase in the volume of global IoT traffic. This exemplary shift initiates a new network forwarding plane. In this letter, we propose and evaluate a Markov Decision Process (MDP) based scheduler to forward diverse IoT Interests to fitting interfaces in an NDN router to fetch Data with less RTT, to meet latency requisites. Simulation results of our MDP model show scheduling Interests to right interfaces reduce the RTT value by approximately 25 − 30% than existing forwarding strategies. The delay is around ∼ 30 ms for higher density of traffic comparatively less than other existing work. Index Terms—Named Data Networking (NDN), Internet Of Things (IoT), Latency, Markov Decision Process (MDP), Delay I. I NTRODUCTION C URRENTLY, IoT has significant growth, as it per- vasively supports a large number of smart objects, connecting diverse applications like smart homes, buildings, healthcare, etc. Unlike current IP, NDN implements a strictly receiver-driven, data-centric communication protocol by ex- change of Interest and Data packets [1]. The named Interest and Data packet has the ability to include additional data on the requested content, its type and application [2]. There is a one to one matching between the Interest and Data packets in NDN to maintain flow balance. A standard NDN node has three data structures Content Store (CS), Pending Interest Table (PIT) and Forwarding Information Base (FIB). The CS acts as a cache to retrieve Data for unsatisfied Interests while PIT monitors the unsatisfied Interests from CS and FIB is basically a table implying details on routing [1]. Generally in NDN, Interests that are unsatisfied by cached Data in CS, create an entry in PIT and further reach FIB for Data retrieval. The simplicity in the forwarding plane has led to a variety of forwarding strategies which have been proposed to achieve adaptive forwarding [3]. There are many available NDN Forwarding Daemon (NFD) strategies [4]: Best route, Multicast, Client Control, Access Router and Adaptive SRTT based Forwarding (ASF). Existing works on forwarding strategies mainly focus on predictive, adaptive and probabilis- tic approaches [5][6]. An MDP-based forwarding strategy is used in [7] to direct Interests to best interfaces, considering a bargain between QoS & OPEX. Although the traditional approaches for forwarding improves the QoS in NDN, issues like delay existed, which does not suit delay-intolerant appli- cations in IoT. Moreover each IoT traffic type needs different treatment while integrating with NDN to efficiently acquire or disseminate Data [8][9]. S.Muralidharan, N.Saxena are with Department of software, Sungkyunkwan University, Korea. e-mail: {shapna2013, navrati}@skku.edu A.Roy is with the System Design Lab, Network Division, Samsung Elec- tronics, e-mail: abhishek.roy@samsung.com Accordingly, in this letter we utilize the MDP based for- warding strategy in FIB to schedule diverse IoT Interests to suitable interfaces and meet the latency requirements. More precisely, in this letter firstly we classify and schedule IoT traffic based on the latency requirements using M/G/1 queuing model using the namespace in NDN to indicate the type of IoT traffic class. Then we have adopted the MDP-based scheduler in FIB to schedule Interests with high priority to interface with high interface ranking to retrieve Data efficiently, to reduce the delay for low latency applications. II. SYSTEM MODEL In the system model, we describe an addition in namespace of Interest and Data packet. To identify the received Interest type, we propose to add the type of traffic class and a timestamp in the namespace as there are no basic restrictions on the name count in NDN [1]. IoT traffic & Network Model: We specify the NDN-IoT network as, G(I , L), where I and L indicate the sensor node set and the communication link. A group of IoT nodes is connected to network A directly to Cache Router by capillary networks. The type of message generated can be event-based (T1), query-based (T2) and time-based (T3) [8]. Mostly, T1 is traffic with low latency requirements, T2 with medium priority and T3 are traffic with no latency requisites. We assume a Poisson packet arrival for various traffic class T at Cache Router with rate λ t [10]. So, the arrival rate of Interest packets at Cache Router for T is also Poisson process with rate λ = ∑ t =3 t =1 λ t . The average time for subsequent Interests arrival at router is 1/λ. The time for service is given by 1/µ and the the system utilization factor ρ t = λ t /µ is given as: ρ = λ µ = ρ 1 + ρ 2 + ρ 3 = t =3  t =1 ρ t (1) We propose an M/G/1 queuing model for diverse traffic types and latency requirements with multiple Poisson arrival streams. The delay for various traffic classes, using Little’s theorem can be expressed as in Equation 2 and 3. E(T 1 ) = E(n 1 ) µ + 1 µ + 1 µ (ρ 2 + ρ 3 ) (2) E(T 2 ) = 1 (1 − ρ 2 ) × µ [1 + ρ 1 − ρ 1 − ρ 2 ] E(T 3 ) = 1 (1 − ρ 3 ) × µ [1 + ρ 1 − (ρ 1 + ρ 2 ) − ρ 3 ] (3) The IoT nodes exchange data to the Cache Router adopting orthogonal subcarriers to avoid packet collisions and transmits at an apparent power to obtain a specific signal-to-noise ratio (SNR) using Additive White Gaussian Noise (AWGN) channel conditions. Using Shannon capacity formula the capacity of subcarrier is calculated considering h t the channel gain at a certain period, W the bandwidth of the subcarrier and η the SNR. It is given by: C s, t = W log 2 (1 + ‖h 2 t ‖η) bits/sec (4)