Models of Computing as a Service and IoT: an analysis of the current scenario with applications using LPWAN

Cloud Computing can be described as a a model to support shared Information Technology (IT) resource accessible on-demand through a network infrastructure [2]. It encompasses a pool of computational resources, like processing, storage, applications, and other services, which are made available through virtualised environments, accessed through the global network infrastructure. This form of computing is increasingly perceived as the 5th utility (after water, electricity, gas, and telephony), which provides the basic level of computing service that is considered essential to meet the everyday needs of the general community [8, 9].

Operationally, Cloud Computing is segmented in three types of services [2]: Infrastructure as a Service (IaaS), where the shared resources relate to computing infrastructure, like Virtual Machines, virtual disks, and others; Platform as a Service (PaaS), where the Cloud infrastructure provides virtualised operational platforms, and; Software as a Service (SaaS) where the consumer has access to shared software running on the Cloud. Each service model has its pros and cons and their adoption relates to the business requirements. For instance, the SaaS model has been widely used by software developing companies to provide solutions through web platform without the need to maintain the complete stack – hardware, OS, HTTP servers, and access; the Cloud SaaS provides all-in-one service accessible through the Web, also including maintenance, support, security, reliability, elasticity, scalability, and others.

The Cloud Computing model introduces a form of distributed computing boosted by a major business prerogative: trade the fixed cost infrastructure around on-premise computing by shared IT infrastructure with variable cost [1]. The business argument is that Cloud clients do not need to afford computing equity, including machinery and physical infrastructure, but instead rent this service from a Cloud provider who also takes care of maintenance, support, depreciation, Quality of Services, infrastructure, redundancy, safety, security, and others. Combine with advances in network communication and scalability of computing power, this model is rapidly becoming the de facto infrastructure for Digital Transformation strategies [10, 11].

However, as business start to migrate workloads to IoT-based solutions, there are many challenges to the Cloud Computing model to support the new computational demands [3, 4, 5, 6]. The term IoT was introduced by Kevin Ashton in 1998 [5] and has been used to designate smart devices with internet connectivity. IoT is flourishing as an important tool to solve problems in several areas of knowledge, such as Smart Cities, Smart Buildings, Industry 4.0, Precision Agriculture, Health, Education, Connected Vehicles, and many others [12, 13, 3]. Each area of IoT application has its specified demands for network consumption, latency sensitivity, physical network layer, distribution, power consumption, and security. Due to this diversity, a large number of companies have developed IoT solutions leading to a heterogeneous and fragmented market [14].

Hence, there is a need to adapt the Cloud paradigm with malleability, distribution, and closer proximity to the data sources. This is the origin of mixed models like Edge Computing, Fog Computing, Mist Computing and others. There is an increased need for localised processing and storage in distributed IoT solutions. Even though usually IoT data is formed by small packages, due to the large number of devices, they generate large data volumes for communication, storage, and processing [15]. Therefore, there is a need to select the proper distributed cloud computing to support the requirements of specific IoT solutions [16].

Telemetry protocols have been incorporated into existing Cloud Computing services to address the issues of data volume and velocity in IoT configurations. To cite the main ones, MQTT [17], CoAP [18], DDS [19], AMQP [20], XMPP [21], STOMP [22] and HTTP19 are example of popular publish-subscribe protocols New challenges in combining these models revolve around issues of synchronisation, configuration, security, vulnerability and others [4, 5, 7]. For instance, more sophisticated security mechanisms require larger computing capacity, such as processing, memory utilisation, and power consumption. Engineers must measure the trade-offs between performance, security, and expected device cost while designing secure IoT devices and deploying distributed Cloud Computing configurations.

With the IoT bringing so many connected devices, some concerns come to the forefront, such as aspects of:

• •

  security, where flaws are frequently reported by data manufacturers [4, 5, 6], system intrusion [23, 24, 25], and others;

• •

  data volume and data flows pose a problem to IoT applications, where it is important to understand the characteristics of the data flow between: (i) small and simple data streams, e.g. coming from e.g. smart metering applications [26], and; (ii) complex and larger data flows , e.g. sources from images, patterns and matrices [27];

• •

  limited or heterogeneous networks;

• •

  energy consumption, which becomes a major challenge when applied to constrained devices [28], often designed to perform with a long battery life, e.g. ultra-long life battery can last for 10 years if the device is properly operated.

Finally, there is a surge of implementations around Low Power Wide Area Network (LPWAN), where devices connect to centralised services through wireless protocols with limited data transmission and radio link.

There are some case where Cloud Computing looks like the ideal option for IoT. This model provides low cost processing and storage, availability, well-known programming resources, and others The concept works well in situations where wither there is good connectivity or no demand for real-time processing. Popular examples are applications of video security in home automation that store streams on the Cloud. Li et al [29] presents the example of Connected Vehicles Cloud Computing where vehicle-bounded sensor devices connect to Cloud-centred infrastructure for services. Figure 1 depicts the solution architecture, composed of:

1. 1.

   cloud layer, providing centralised processing and storage;

2. 2.

   service consumers layer, which in the case of IoT applications service consumers are sensors.

Sensors must be connected either directly through Application Program Interfaces (APIs) or through a Protocol Gateway. All services are performed on the Cloud infrastructure, including processing, storage, and any other add-value service. If location information is needed, the gateway (or sensor) must include location information such as pre-configured or from a GPS, as the central service lacks awareness of location [30].

This architecture presents challenges to support IoT-based applications mainly due to volume, latency, heterogeneity, security, and others. Basu et al [31] list some important challenges, such as: data segregation, data location, data incomplete data, monitoring and data logging, problems associated with the security of Virtual Machines and their environment, and even natural disasters. Subramanian et al [32] enlist confidentiality and integrity among the main security challenges.

Other requirements are authentication, auditing, and legal security requirements. The latter is important because the data is subject to regulations in some countries. Sha et al [12] describe the challenges around heterogeneous network technologies, privacy, large scale of systems, and management of trust. Stergiou et al [33] corroborate to the narrative, listing key challenges around heterogeneity, performance, reliability, big data and monitoring.

Fog Computing is a model of distributed Cloud Computing designed to cope with the growth of IoT environment and issues of latency inherent to this configuration [34, 30]. Yi et al [7] postulate that this model will be the internet of the future due to the number of existing devices and the arrival of more devices in the market.

This model is organised in three layers, as depicted in Figure 2:

1. 1.

   Cloud with the classic CC model and centralised computing;

2. 2.

   Fog Nodes are central part of the architecture; these can be either physical components or virtual components and are tightly coupled with smart devices or network access nodes, providing computational resources to these devices.

3. 3.

   Sensor Nodes providing the data sources in the edge layer.

This model is inherently distributed geographically, providing resources near the data sources where they are usually most necessary. For example, in case of IoT and wearable devices, impeding restrictions of processing and storage in these devices demand for Fog Nodes to deliver situated resources.

However, the multi-layered nature of this model increases its implementation complexity. One direct impact of this extra complexity is on security. Yi et al [7] describes the challenges of Fog Computing around the choice of virtualisation technologies to be applied, along with the issues of latency, network management, security, and information privacy. Luan et al [30] includes challenges around application development of applications, scalability, and distribution.

DSouza et al [1] mentions that Fog Computing’s multi-level collaboration brings a new set of problems around identity management, resource access management, distributed decision, dynamic load balancing, quality of service and security. Vaquero et al [15] mentions that in order for Fog Computing to become reality, developers must first solve its many challenges around synchronisation, device discovery, management, security, standardisation, accountability, billing of mobile applications.

Mist Computing (MC) plays an important role in the migrating computation power to the systems’ edge. This model extents Fog Computing by adding an extra layer placed closer to the edge, below the Fog Nodes [35]. Figure 3 depicts the architecture:

1. 1.

   Cloud Computing provide centralised services, when needed, such as monitoring, updates, central data repository, offloading processing, and others.

2. 2.

   Fog Computing provides distributed capabilities and group control over the mist nodes, bridging with Cloud Computing services when required; for instance, Fog Nodes may provide resources like regional storage, processing offloading, monitoring, software updates, and others.

3. 3.

   Mist Nodes are located at the border, providing local resources and processing, with connectivity with their peers and requesting resources from them.

Mist Nodes are usually implemented through devices that offer basic computational power, such as Arduinos, Nodemcus, and other microcontrollers and microchips. A practical example of Mist Computing implementation exist in automated vehicles where the multiple elements are connected to a central car unit (Fog), which can connected to other cars for collaboration and collective intelligence.

The key concept is to promote interaction between Mist Nodes as much as possible, refraining from utilising centralised services or devices. The architecture can be conceptualised as a model in which network edge devices have predictable accessibility and provide their communication and computational resources as a service [36]. Mist Nodes can distribute software processes to run on service providers on their own. That is, Mist Computing favours a model of computing “a hop away” [37].

Mist Computing brings computing power deeper into the edge, embedding processing in microcontrollers and System on Chips [38]. This model can provide a flexible environment for execution of customised programs [36]. However, Mist Nodes do not have the computational power of a Fog Node so they must be used in complement. In situations where applications demand processing or storage, these requests can be offloaded to Fog Nodes, which in turn can relay to Cloud Computing.

This architect can be leveraged to extend the capacity of constrained devices, like mobile sensors, and promote fast processing at the very edge of the system. Being an integral part of the edge, Mist Computing provides the lowest latency possible for an IoT application. However, it suitable only for a restricted number of application scenarios considering the constrained computing capabilities and implicit distribution.

Preden et al [39] describes the challenges of Mist Computing related to communication complexity and self-management, as a result of the dependence of a central component. For Yogi et al [40] list as challenges: low storage capacity, limited bandwidth, and resistance from solution developers to adopt the model. Vasconcelos et al [41] argues that the challenges are common to Fog Computing, and the most characteristic challenge is related to the complexity brought by the dynamic topology. Suarez-Albela et al.[42] list challenges related to security, as the use of security mechanisms such as encryption demand a require energy consumption by the devices.

Mobile Cloud Computing (MCC) is about the provisioning of data and storage closed to the mobile user. In its most elementary form it refers to an infrastructure where both data processing and storage take place outside the mobile device [7]. The concept of mobile cloud relates to pervasive devices sharing their support services and heterogeneous resources, such as network band, processing, content, and others [43]. The architecture contains three layers as depicted in Figure 4:

1. 1.

   Front-end layer, where either user interfaces or services request applications are located, usually running on mobile devices; this layer works by requesting service from external providers, through offloading processing and storage.

2. 2.

   Middle layer, promotes offloading access to computing power hosted on servers, wireless network, or cloudlet [44].

3. 3.

   Cloud layer, provides the back-end infrastructure to respond to all requests.

Mobile Cloud Applications can transport computing, energy and data storage out of phones and into the Cloud. This allows to create a new range of solutions not only for mobile phone applications but also a range of other solution niches [45].

In order to prevent delays in processing the data due to latency or network bottlenecks, the network segment between the front-end and the middle layer must support high throughput. Moreover, the communication link between the middle layer and the cloud is usually the Internet, thus subject to intermittent connections, latency, and security issues. Hence, real-time applications must be processed in the middle layer. The element responsible for bringing cloud capabilities close to the mobile device are the cloudlet [43].

Au et al [46] argues that the key challenges for this model are related to data and authentication, including: authentication of mobile users and devices, security in data communication and storage, data integrity, data search, and secure data sharing. Leppanen and Riekki [47] enlist as challenges offloading, scheduling, monitoring, resource tracking, context awareness, and remote service availability. Challenges and issues of heterogeneity in Mobile Cloud Computing are largerly discussed in [14].

Sekaran, Vikram and Chowdary [48] present the issue of security and Distributed Denial of Services (DDoS), and describe different ways to prevent these attacks in MCC. Noor et al [49] list as the main challenges security, privacy, bandwidth control, data transfer, data management, synchronisation, energy efficiency and heterogeneity.

Mobile Edge Computing (MEC) utilises the infrastructure based on mobile networks to provide connectivity to edge devices. This architecture has the purpose of bringing computing resources, mainly processing and storage, near to the data sources. Services are deployed directly at base stations or in smart cells, as for instance Femtocells, Picocells, Nanocells, and others. This approach relies on the infrastructure provided by mobile phone operators and it may incur in issues of coverage, high costs of mobile data, and others.

The architecture is implemented in three layers, as depicted in Figure 5:

1. 1.

   edge, usually composed of IoT , mobile computing, and wearable devices;

2. 2.

   servers, providing located computing resources aiming at low latency and quick response for service requests;

3. 3.

   cloud, providing centralised processing and storage.

MEC brings some relevant benefits such as ultra-low latency, high bandwidth, real-time access to radio network information, and location services [51]. For the latter, unlike other models, MEC allows to identify the location of the data source, thus providing additional support for security, locality, and auditing e.g. in case of regulated installations. Moreover, ultra-low latency favours applications that require quick response in decision-making.

The devices located on the edge of the MEC must have components that allow them to access the cellular network, either via macrocells i.e. antennas and other classic devices or smartcells, which are eager to increase the reach of cellular networks, by expanding their capacity to provide connectivity, specially to o rural areas.

This model is becoming a hot topic with the multiplication of MEC service providers and the arrival 5G technology, in solutions to support communication, computing, control, and content delivery [52]. One the key benefits of this model for IoT solutions is better coverage, specially in urban areas. This will allow for like Smart Cities and Smart Buildings that can directly benefit from MEC and 5G.

The server layer brings mobile users close to the benefits of Cloud Computing around elasticity and virtualisation. In the MEC model, this is implemented upon virtualisation platforms like Network Functions Virtualisation (NFV), Information-Centric Networks (ICN) and Software Defined Networks (SDN) [52].

NFV promotes the virtualisation of network functions, making them work like in the Cloud Computing model. This is implemented either through dedicated hardware or off-the-shelf devices [51]. The objective of virtualising network functions is towards the cost benefits of shared infrastructure, preventing e.g. capital costs and operational costs related to site allocation, cooling, maintenance, and others. The functioning is based on three concepts: virtualisation, orchestration, and turning all functions into software. Common examples of virtualised functions include firewall, DHCP, NAT, and others

SDN virtualises the network by decoupling control plane from data plane [51]. As an analogy, NFV visualises network functions, SDN. The goal is to understand the allocation and performance of routes, define new ones, connect physical and virtual services, define policies, allocate IPs, allocate bandwidth, and ensure connectivity. SDN facilities the process of creating NFV by supporting the configuration of the connection of functions in the network.

Ahmed and Rehmani [53] list as challenges the fast development of services at a cost efficient rate, optimised resource utilisation, facilitate the migration of existing application, and security. Vassilakis [54] mentions the need for security and privacy solutions specific to this model and the possible coexistence with unreliable nodes. Beck et al. [55, 56] defines some metrics that need to be met to allow utilisation in some areas, such as energy consumption, delay, bandwidth and scalability. They also mention challenges around scalability, mobility awareness, and utilisation awareness by embedded applications.

Varghese et al. [57] presents a list with five challenges: (i) ability to provide general purpose computing at the edge nodes; (ii) node discovery; (iii) task partitioning and orchestration; (iv) balance demands, processing, and Quality of Service (QoS) and Quality of Experience(QoE); and (v) safety. Roman, Lopes and Mambo [58] list as challenges: infrastructure, virtualisation, resources and tasks, distribution, mobility and programmability.

There are two trendy technologies emerging in the market that are worth mentioning at this point: Superfluid Cloud and Cloud Radio Access Network. We acknowledge that this list is far from exhaustive and other commercial and research solutions emerge frequently in the market, but included a brief description for the reference.

Superfluid Cloud [59] is a multi-tenant model where virtualised services based on software execute on commodity hardware, being shared and deployed across the network. The idea is to apply low-cost devices with devices with significant computational power, such as System-on-Chip, Cubieboard, Raspberry Pi, and others. The most important characteristics include [51]: recursion; scalability; separation between state and processing; support for very small VMs; support for Extended State Finite Machines (XFSMs), and; on-the-fly monitoring.

Cloud Radio Access Networks virtualise some important functions of modern telecommunications architecture, lowering the cost of deploying and operating mobile networks [60]. Demestichas et al [61] list challenges around multiple perspectives of society, economy, users, and operators. There are also issues on normalisation in order to have a cohesive, inclusive, and sustainable structure. These challenges relate to wireless communications in general and can be mitigated through C-RAN’s centralised structures.