Page 82 - ITU Journal Future and evolving technologies – Volume 2 (2021), Issue 2

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 2

tems using TDMA, some work allocates time in each time rapidly, and conversely [55]. So these two resources are

slots for each device proportionally to the data they need involved in QoS requirements, such as delay and energy
to of load [79, 52]. For OFDMA, subcarriers are allo‑ consumption, and jointly allocating them lead to more ef‑
cated [57, 83]. The promising Non‑Orthogonal Multiple icient results [62].
Access (NOMA) method, suitable for 5G, allows sharing Server selection The server selection for a task can also
subcarriers between multiple users instead to at most be considered as a resource allocation. Matching MEC

one, like in OFDMA. So if the system uses NOMA, the re‑ nodes with tasks is relevant because MEC nodes may

source allocation scheme had to assign the subcarriers possess heterogeneous capacities, in terms of quantity

to multiple users [71, 50]. Some work also model com‑ as well of quality, and it more or less a task need [65,

munication resources abstractly to be applied on differ‑ 70]. Plus, there is a trade‑off to consider between com‑

ent types of systems. A point to consider when we al‑ putational time and network delay, depending on servers’
locate communication resources is the interference. The workloads, their distance from devices and their channel
intra‑cell interference is usually ignored to sub‑channels quality. For example, it may be worthier to assign a task

assignment [49, 62]. Inter‑cell interference makes the re‑ to a farther server but which is less busy [65].
sourceallocation morecomplex as it adds dependence be‑
tween users’ uploading rates [62]. Some work ignores 4.3 Goals

this inter‑cell interference as they postulate that cells are
farenoughfromeach otheror haveorthogonalbandwidth The goals of the resource allocation scheme depend on

allocation [49]. However, it can be interesting to con‑ the use cases and the applications considered by the

sider it for networks where these interferences are highly work. They can be more adapted to latency sensitive

probable, like ultra‑dense MEC networks. In addition, as applications by minimizing tasks’ completion time, or it
MEC may possess heterogeneous Radio Access Technolo‑ MEC systems with battery‑powered devices by minimi-
gies (RATs), it raises the interesting problem of choosing zing their energy consumption. The goal can be

the right RAT to serve a device at a given time for a given tuned with weight in the objective function. It can aim to
task. For instance, Hsu et al. [72] consider the licensed 5G prioritize some devices [62] or some aspect of a
and the unlicensed NR‑U. Indeed, each RAT has its own multi‑objective problem [49], like giving more weight to
characteristic, like coverage, mobility support, data rate energy consumption rather than latency.
and so on. All of this may in luence the delay, the energy
consumption and the quality of service. They can also in‑ 4.3.1 Energy‑aware parameters
cur additional costs, like 5G instead of the generally free
Wi‑Fi. Finally, next generation emergency [84] and public In mobile edge computing networks, mobile devices are
safety [85] communications are challenges to incorporate battery powered. Thus minimizing their energy con‑

in MEC resources allocations. sumption is crucial to maintaining the user’s quality of

Computational resource Instead of the cloud, MEC sys‑ experience [62] and preserve autonomous devices’ bat‑

tems have limited computational resources. Thus they tery to let them complete their tasks. Some work con‑

are critical resources we need to allocate ef iciently. It is siders the overall energy consumption, e.g., from the lo‑
even more the case with many mobile users or extremely cal computing to the of load computing [57]. Other work
limited edge resources, as it is often the case in mission‑ considers only the device’s energy consumption as is it as‑
critical applications. If the computing resources are badly sumed that MEC servers have reliable power sources [47,
allocated, important devices’ tasks may be unprocessed 49]. However, in mission‑critical applications, servers

on time and the overall system can be congested. Compu‑ can be battery powered, like embedded on UAVs or in

tational resources may be CPU cycles per seconds [55, 80, buses. They may have more energy at their disposal than
62] or CPU cores [56]. For UAVs‑based MEC, some work end devices; nonetheless, their energy budget is limited.
allocates the number of of loaded bits to the UAV [86, 87]. Moreover, the energy consumption of the overall system
However, recent work considers allocating CPU frequen‑ is always important to minimize the application’s energy
cies instead because it seems to reduce energy consump‑ impact. The energy consumption for a task is often calcu‑

tion further [78, 88, 89]. To reduce the latency further lated as = ⋅ , where F is the computing capacities
and help MEC servers, some works also consider the use of the device as CPU cycles per seconds, c represent the
of some spare computing power on certain devices that number of CPU cycles required to inish the task, and
can use it to assist other devices. The devices with enough are constant that depends on the device’s chip architec‑
−
−
resources communicate directly with requesting devices, ture [54, 62]. is often 10 26 or 10 27 and is 2 [59, 70,
called device‑to‑device [90] or machine‑to‑machine [71] 53, 49]. So the computing capacities in luence the task’s

communication. process time but also the energy consumption. Therefore,
Joint communication and computation The communi‑ there is a trade‑off between energy consumption and
cation and computation allocations affect each other. Re‑ execution time to consider. This trade‑off can be

gardless of how much a task is given a certain amount adjusted with a weight factor in the optimization goal to
of communication resource, if it does not have enough it the application needs, having a low energy

computing resource, the task will not be processed more consumption or a reduced latency [91, 59]. It can

68 © International Telecommunication Union, 2021