Representación de los datos alfanuméricos

In our simulations, as well as seeing the savings made by offload, we are also interested in what activities (computation or communication) the energy was spent on before and after the offload. Fig. 4.8 includes two stacked bar plots. The one in the background shows the energy distributions of the original workflow. The second plot, in the foreground, shows the energy distributions of the offloaded workflow. The top section of both plots indicate the share of energy that is spent on communication. The data is grouped so that on the very left is the data gathered from workflows that are offloaded only to take advantage of the fast processing speeds of the cloudlet (i.e. with poor network bandwidth). On the very right are data from workflows that are offloaded only

10:0 9:1 8:2 7:3 6:4 5:5 4:6 3:7 2:8 1:9 0:10 0 0.2 0.4 0.6 0.8 1

Smartphone to cloudlet processing speed ratio

Energy

Cost

Distribut

ion,

%

Before offload computation Before offload communication After offload computation After offload communication

Figure 4.8: Energy distribution before and after offload. Offloaded workflows are proportionally more reliant on the network.

to eliminate communication costs (slow cloudlet speed). We can clearly see that in all groups, the share of energy spent on communication has been increased after offload. This is a clear indication that an offloaded workflow is proportionally more reliant on network connectivity than its original form.

4.4

Summary

In this chapter, we presented our approach to managing a mobile workflow over its supporting platform in an energy- and time-aware manner. With a model which reflects the software and hardware characteristics of the scenario, we developed a heuristic algorithm to build and update the offload plan dynamically based on the time and energy constraints of the workflow. Variations of the objective functions are also discussed together with optimisation of the algorithm.

A series of simulation studies concludes that:

1. When no code repository is available at the server side, a large executable size invariably generates a negative effect on a workflow’s offload-ability.

4.4 Summary

tasks are concentrated on a small number of smartphones.

3. Energy savings can be found easier on workloads that are not on the workflow’s critical path, so even when offload is proven not to be preferable by the time constraint, savings can still be made in the workflow’s overall energy consumption.

4. The significance of the savings brought about by offload follow a reciprocal relation to the hardware metrics.

Bandwidth Dependency and

Allocation in Mobile

Service-Oriented Networks

Bandwidth is another influential factor to the QoS of mobile applications and services alongside energy. As we have demonstrated in the previous chapters, the bandwidth variable plays a pivotal role in the management of mobile cloud computing platforms. When the services requested by mobile application workflows are distributed over a network of cooperative mobile smart devices and clouds, the question arises as to which service should be allocated with how much bandwidth and when in order to satisfy service demands? Moreover, how to adjust the bandwidth allocation to accommodate changes in service demands whilst maintaining service QoS? Furthermore, the mobility of smart mobile devices brings forward the challenge to determine how changes in mobile network conditions affect the bandwidth requirements of interacting services.

To answer these questions, in this chapter, we investigate the resource management aspect of mobile cloud computing platforms, more specifically on modelling the bandwidth dependencies between interactive components of mobile application workflows. We generalise the bandwidth allocation problem in mobile cloud computing platforms to that of generic mobile wireless net-

5.1 Mobile Service-Oriented Networks

works which we refer to as Mobile Service-Oriented Networks (MSON)s, so that the model is applicable to a wider scope of problems. We assume all computation nodes are mobile devices and don’t specifically include cloud nodes in our model. This generalisation does not alter the structure of the model since we wish to quantify the bandwidth requirements of each computa- tion nodes and whether the node is mobile or cloud does not affect this value. We give further definition of an MSON in Section 5.1.

In Section 5.2 we give introduction to the Leontief I-O model which is th foundation of economic studies. Then in Section 5.3 we adopt and extend on the analytical framework of the Leontief Input-Output model and develop a network I-O model to describe the bandwidth dependencies within an MSON. Various factors such as bandwidth, latency, service demand and costs are accounted for in the model. A set of equilibrium equations are derived to produce the bandwidth requirements of mobile devices.

Based on the Network I-O model, first a cost-based bandwidth allocation scheme is developed to maximise service benefit in Section 5.5. Next, a set of adaptive bandwidth allocation strategies is also proposed to accommodate changes in service demands and network conditions while minimising the overall impact on application workflows in Section 5.5. Results from simulation studies are presented at the end of each section to demonstrate the effectiveness of the proposed methods.

5.1

Mobile Service-Oriented Networks

In this chapter, we assume that mobile (cloud) application workflows are manifested as dynamic compositions of services located on mobile devices. We refer to the underlying network structure as a Mobile Service-Oriented Network (MSON).