Page 16 - ITU Journal Future and evolving technologies Volume 2 (2021), Issue 5 – Internet of Everything

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 5

Table 1 – Literature contributions

Standard Literature Paper contribution
[36, 37, 38] Proposal of a revenue maximization framework based on data information elab‑
orated locally on the users’ devices, avoiding the typical privacy concerns of the
other approaches.
[40, 8, 39, 20, 41, 42] Contextualization of the application of the FL to the VFs deployment problem, by
exploiting the FL framework to properly predict the application network demand,
in order to maximize the SP revenue.

Table 2 – Main symbols

Notation Description 0, = 0,
( ) = { (1)

CN Computation node , + , > 0,

,
VF Virtual function
FL Federated learning in which , , are real valued parameters whose
,
,

SRB Storage resource block value changes on the basis of the request type.
SR Service request Similarly, the provision cost for providing SRs of type

S Number of SRBs per CN follows the rule [46]
U Cloud SRBs
ECN Edge computing network ( ) = { 0, = 0, (2)

High priority requests , + , > 0,

,

ℳ Low priority requests
Time deadline where , , are, also in this case, real valued pa‑

,
,
Number of req. demanding for service rameters.
Number of req. demanding for service Moreover, for each service type with high priority, the SP
( , ) SP revenue for the high priority req. revenue results ruled by the following relation

( , ) SP revenue for the low priority req.

Service accomplishment time ( , ) = log(1 + ) , (3)

,ℎ Waiting time on the CN
, Waiting time on the cloud
with = | − |, where is the number of SRs for

1
any CN to the cloud . Furthermore, we guess that the ECN which has been respected. Then, the SP revenue for the
low priority SRs is given by
is able to support different high priority service types,
which are characterized by different provision costs and log(1 + )
selling prices. Each service type ∈ has associated a ( , ) = , (4)

lev expressed befor which

the type service accomplishment has to be completed. where is the number of SRs among accepted by the

In addition, we consider the presence of ℳ service type network for their service. Hence, the SP revenue, corre‑
requests with lower priority and without any time dead‑ sponding to the provision of the ‑th and the ‑th service
line constraint. The number of requests belonging to this type, can be expressed as
class is indicated hereafter with , with ∈ ℳ.

Periodically, the SP updates the service demand and we ( , ) = ( , ) − ( ), (5)

assume that any new request does not arrive between two
SP updates. and
Let be the number of SRs demanding for service We ( , ) = ( , ) − ( ), (6)

that originat b that
respectively.
an EU requires only one SR. Therefore, as a direct conse‑
Both the SRs with high and low priority, in order to be
quence, hereafter we assume interchangeable the SR and
accomplished, require the presence of a VF in set which
EU Then, as regards the SP, the provision of a ser‑ has to be preliminary loaded on at least one CN of the net‑
vice has a cost mainly depending on and following the work or on the far cloud. The loading process requires the

model given by [46] CN or cloud availability in terms of SRBs, since each VF
∈ requires a number of SRBs, different for each VF.
1 We have assumed that the connection towards the cloud is performed
throughout the CN nearest to the SRs needing computation. Conse‑ Consequently, the time required for the service accom‑
quently, the communication latency cost between SRs and their nearest plishment (TSA) of a generic SR , independently by its
CN has no impact on the overall SR completion time and hence it has priority, is given by
been neglected in de ining (7).
4 © International Telecommunication Union, 2021