1.5 Organisation of The Thesis
2.1.4 Comparison of Mobile Cloud Architectures
In this section, the different mobile cloud approaches are compared in Table 2.1. The evaluation is based on the following properties, which have been chosen due to their effects on the Quality of Service of mobile applications:
Latency: This property evaluates how fitting is the system for latency-critical
applications. In some mobile applications such as Augmented Reality, users request low latencies to avoid having slow frame loading, which affects the performance. As shown in Table 2.1, MCC solutions that offload to Remote Cloud servers have a high latency which does not make them ideal candidates for low latency applications. The delay-sensitive mobile tasks are ideally offloaded to the cloudlets or nearby devices. On the other hand, differences between Edge Computing and Fog Computing are dependent on the location of the edge servers to the mobile user as they are confined to the edge infrastructures.
Distance: This depicts the physical distance of the infrastructure providing the
2.1 Mobile Cloud Architectures 19
number of network hops that are being introduced further the mobile device gets to the servers. Nearby local devices are usually within 10-20 meters away, while the Cloudlets and edge servers can be as far as a kilometre to Host Mobile Device.
Deployment: The cloud service providers such as AWS and Azure are usu-
ally the sources of Remote Cloud servers. To resolve latency issues, fixed surrogates can be deployed in the vicinity to be a one-hop away from the mobile device in the form of Cloudlets. These can also be in the form of macro-datacenters, small clouds, femtocells, etc. MEC servers are deployed in Radio Access Networks which can provide fast services to mobile users.
Computation Power & Storage Capacity: The abundance of resources at
cloud data centers and the elasticity feature of Cloud Computing allow for ample computation power and storage capacity. Even though cloudlets and nearby devices solve the latency issues of far clouds, their computation power is rather limited. Edge and Fog Computing use a distributed set of powerful servers deployed in the network edges to provide powerful services to mobile users.
Communication Medium: Mobile devices connect to different service providers
using different wireless technologies. Since the public cloud infrastructure is accessed through the Internet, both Wi-Fi or cellular technologies can be used to establish connections between them. Cloudlets are deployed in Local Area Network (LAN) so they can be connected to devices in the local network. Traditionally, Bluetooth has been the most popular choice for Device-to-Device (D2D) communications. However, there are multiple new approaches to achieving nearby peer connection establishments, such as WiFi-Direct and ZigBee [19]. The Edge servers are mainly deployed in network servers, so cellular technologies such as 3G/4G are the common network interfaces to connect to them.
Architecture: The architecture tier is a physical structuring mechanism for
the system infrastructure in which the computation levels depend on the existence of devices and communications. The most common architecture is when the mobile device is only connected to a single Remote Cloud server that forms a 2-tier mobile-cloud architecture. Cloudlets are used as middle managers to form a 3-tier mobile-cloudlet-cloud architecture for enabling the cloudlet to be closely connected to both the mobile device and the cloud. The MAC devices are normally in the same architecture tier. Meanwhile,
MEC and MFC do not have a standard architecture so devices in different cloud resource levels can be included such as smart routers, nearby mobile devices, forming a multi-tier architecture as shown in Figure 2.1.
Availability: Service availability is one of the most important factors affecting
responsiveness and energy consumption in mobile device augmentation scenarios. Cloud providers guarantee high availability of services through a Service Level Agreement. Although cloudlets are deployed close to mobile users, in cases of service churns and multi-user requests, the limited server might not be able to serve all the users in the same way as remote cloud servers. As for MAC, because of the mobility feature of nearby mobile devices, it makes them highly volatile and could become non-available on the fly [93]. MEC and MFC solve the issue of mobility using horizontal and vertical handoff strategies to avoid service disruptions for mobile users [86].
Use Cases: The applications running on Cloudlets and Nearby Mobile Devices
are typically time critical and require very low computing and communication latency in an environment with limited bandwidth, and limited computing power. For example, applications such as Linpack (computation-intensive), 3D Car Racing (interaction-intensive) and Chess (computation & interaction- intensive) [147]. MEC is actively used in smart city planning and video surveillance scenarios [86] while MFC conducted studies are mainly targeted towards IoT applications [17].
Operators: Cloudlets are mostly deployed by an individual or a local business
such as a coffee shop owner [49]. Since MAC is developed in an ad-hoc fashion, it can be performed locally by an individual or a number of individuals within an organization. The network providers have added powerful edge servers in their base stations to enable MEC [88]. Fog nodes can either use the edge servers in base stations or cloud servers in cloud service providers [158].
2.1 Mobile Cloud Arc hitectures 21
Table 2.1 Comparison of mobile cloud related paradigms
Property Core CloudletMCC MAC MEC MFC
Latency High Low Low Low Relatively low
Distance Far Close Very close Close Relatively close
Deployment Data centers Fixed surrogates Nearby devices Network edge
(RAN) Fog nodes
Computational power & Stor- age capacity
Ample Limited Very limited Fair Ample
Communication
medium WiFi/Cellular WiFi WiFi/Bluetooth Cellular WiFi/Cellular
Architecture 2-tier 2-tier or more 2-tier 2-tier or more 3-tier or more
Availability High Average Low Average High
Use cases Social networking,
health care VR) appsImmersive (AR & Disasterlief, privacy-re- preserving local processing
Video surveillance, video caching, traf- fic control, health monitoring, AR
IoT, Connected ve- hicles, smart city, smart delivery
Operators Cloud service
providers local businesses self-organized Networktructure providersinfras- (RAN-based)
Cloud service providers and net- work infrastructure providers