Page 38 - ITU Journal Future and evolving technologies – Volume 2 (2021), Issue 2

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

In Fig. 1 the network topology is illustrated. Solid lines 4. CLUSTERING AND CACHE PLACEMENT
with solid arrows indicate D2D communication. D2D OPERATIONS
links between HUs are required for intra‑cluster opera‑
tions. Four of the OUs are highlighted to depict some par‑ Proposed ABD2D and DBD2D procedures are envisaged
ticular cases. 1) OU1 is not close enough to a CH and only to work in conjunction with DBCA and HCA that were
a nearby IHU serves requested content. 2) OU2 neither originally proposed in [34]. Therefore, in the scope of
has a CH nor an IHU in its D2D range for the correspond‑ this paper, we evaluate the performance of these algo‑
ing time slot. 3) OU3 is in coverage of two CHs. 4) Simi‑ rithms once again and investigate the adaptation of pre‑
larly OU4 has a link to a CH and is able to exploit content proposed algorithms and currently proposed procedures
from cluster members. Moreover, dashed lines with solid as a whole. Shortened descriptions of the algorithms are
arrows indicate a cellular link and a dashed line with the given below.
thin arrow indicates the control plane interface between
Distance Based Clustering Algorithm (DBCA) aims to
the BS and ProSe Server. The role of these links is de‑
scribed in the following sections. cluster nearby HUs and pave the way for facilitating col‑
laboration among cache enabled devices by providing
content diversity. In our clustering approach: 1) All clus‑
3. OFFLOADING FORMULATION ter members are close enough to establish direct D2D
links between each other, this enables intra‑cluster coor‑
For the sake of completeness, the problem of of loading dination. 2) Cluster members act in a hierarchical struc‑
maximization, originally formulated in [34] is given in (2). ture, the HU that has the most number of OUs in its D2D
We assume that there are ordinary users, helper range is designated as the Cluster Head (CH). Cache place‑
users and contents. Here is popularity of content ment is managed by the CH and cache capacity of the clus‑
for OU , which is Zipf distributed and uniform among all ter is considered as a whole. 3) Each HU can be a member
OUs. Decision variable ℎ takes binary values and indi‑ of only one cluster; this constraint prevents con licts in
cates delivery of content to OU from HU ℎ. Parameter caching. We assume that the clustering operation is man‑
stands for content size. Another binary variable ℎ in‑ aged by the D2D ProSe server, which receives the neigh‑

dicates that content is cached by helper ℎ. The sum of borhood information from HUs as input, as in [26].
ℎ along ℎ is limited to cache capacity . Moreover, ℎ

stands for the SINR based D2D eligibility between HU ℎ Cache placement operation is managed by the CH relying
and user . Accordingly, (5) enforces that an OU can re‑ on the Hierachical Caching Algoritm (HCA) [34]. This
ceive help from HU ℎ, in terms of content , only if HU ℎ algorithm is managed by the head of each cluster. The
caches that content and HU/OU meets the SINR threshold algorithm accepts cluster information, content popular‑
condition. ity ( ), number of contents ( ) and cache capacity of

HUs ( ) as inputs. For each cluster, initially total cache

capacity is calculated and cluster level caching layout is
= {∑ ∑ ∑ ℎ (2) determined as to how many copies of each content will
}
=1 =1 ℎ=1 be cached. Then HU‑level cache placement is determined
according to the cluster’s hierarchical order. Resultant
s.t. cache placement may include duplicate copies of popu‑
lar content in multiple cluster members, or some content

∑ ≤ , ∀ℎ = 1, … , (3) may not be cached at all. As an exception, HUs that are

ℎ
=1 not members of any cluster are classi ied as isolated HUs
(IHU), cache the most popular content and serve OUs in‑
dividually. Content to be cached is requested from the BS.

∑ ℎ ≤ 1, ∀ = 1, … , , = 1, … , (4)
ℎ=1 In this work, we assume that clustering and cache place‑
ment procedures are repeated periodically in order to
cope with mobility. The robustness of the algorithm pair
ℎ ≤ , ∀ = 1, … , , = 1, … , , ℎ = 1, … , (DBCA and HCA) against mobility was investigated in
ℎ ℎ
(5) [44]. A reasonable update period of 10‑15 seconds re‑
sulted in a small performance loss (around 4 − 5%).

The formulation described above maximizes of loading 5. PROPOSED ADVERTISEMENT AND DIS‑
for a given (static) network topology and disregards mo‑ COVERY PROCEDURES
bility. Besides, it assumes perfect user/content discovery.
For a mobile network, this optimization problem has to be As mentioned in Section 1, previous studies on D2D
solved periodically with suf icient frequency and it has a caching (including [34]) mainly overlooked D2D user and
prohibitive complexity. content discovery. In this section, the proposed content

24 © International Telecommunication Union, 2021