Page 88 - 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

Concerning aspects such as secure channel establish- structures of both the CLO that wants to establish a friend‑
ment, COG‑LO identi ied a series of shortcomings to the ship and the data structures of all involved friends (since
cornerstone technologies that facilitate secure informa‑ the social relations are bidirectional).
tion exchange over the Internet, namely the Public Key As usual in traditional distributed deployments of sys‑
Infrastructure (PKI) and associated X.509 certi icate stan‑ tems like the SIoT platform used in this project, different
dard [30]. ically, as it has been recently shown, servers are used to share the load of traf ic and compu‑
PKIs are exposed to risks due to errors or breaches in‑ tation. Each server can be con igured to work following
volving Certi ication Authorities (CAs), resulting in unau‑ a full replication or a partial replication scheme, or even
thorised certi icates being issued and compromising thus
as a totally independent system with non‑replicated data.
the security of the corresponding end users. In light of
Using the irst approach all the data is replicated or copied
the above, COG‑LO adopts a novel blockchain‑based solu‑
to all the participating nodes in the cluster. Otherwise, us‑
tion enabled by the Hyperledger [27] family of technolo‑
ing the second approach, the entire data is split equally
gies, namely the Hyperledger Indy and the Hyperledger into partitions and is stored in the participating nodes,
Aries frameworks in order to establish secure cross‑ thereby creating a distributed storage of data. The to‑
organisational communication. The aforementioned so‑ tal storage space depends on the total memory available
lution, being based on the blockchain technology, inher‑ across the peer. As shown in Fig. 5, the replicated mode
its inevitably its advantages. The solid basis of the pub‑ allows the speed‑up of the discovery process, since the in‑
lic append‑only log (past logs cannot be changed unless formation is immediately available in the peer from which
the blockchain is subverted by a dishonest network ma‑ the search is being performed. While this approach has
jority), eliminate the single‑point‑of‑failure issue and en‑ bene its in terms of time, it also requires an increased use
ables rapid reaction to identity revocations since DIDs can of resources.
be validated on the distributed ledger.
The optimization performed by the CA is based on the
graph representation of the CLOs network. Upon the oc‑
4. PERFORMANCE ANALYSIS currence of a disruption event (e.g., an ad hoc order, a
tr ic event etc.), the graph gets pruned by SIoT to in‑
The fundamental aspect that is used to evaluate the per‑
clude only CLOs in the vicinity of the event. The size of
formance of the COG‑LO framework is the time consump‑
the graph, i.e., the number of vehicles included in the op‑
tion of the algorithms implemented within the Social In‑
ternet of Things (SIoT) and the optimizer. timization impacts greatly the response time.
More ically, the scalability of these components is The performance of the optimization algorithms is pre‑
addressed by observing how the computational time for sented in Table 1. It clearly shows the importance of graph
object digitalization varies as the number of objects in‑ pruning to achieve real‑time responsiveness to disruption
creases, how the social graph ing time varies when events. Table 1 shows the performance time for optimiza‑
a friend must be discovered as the size of social graph tion processing on pruned graphs. An optimization algo‑
varies, and inally by the complexity of the optimization rithm uses exact methods, with linear solver, where us‑
algorithms in large environments, where the problem of ing a large number of CLOs exponentially increases pro‑
a large amount of data and variables to be analyzed must cessing time. For optimization processing, the total in‑
be faced. frastructure graph is clustered into regional representa‑
The Social Internet of Things (SIoT) architecture consists tion with graph sizes of 300‑500 CLOs (postal of ices, ve‑
of various SIoT clusters. Each cluster is implemented hicles, parcels). The processing time in Table 1 clearly
using Apache Ignite to support a memory‑centric dis‑ shows that without pruning the graph and omitting the
tributed database, caching and processing platform [17]. number of CLOs included in event handling, the system
would not be able to create real‑time responses.
Each SIoT peer can automatically discover each other in
order to create a cluster or to browse another peer’s so‑
cial graph. In the irst performance study of the SIoT, the
Table 1 – VRP optimization response time, based on the number of CLOs
response time for the creation of a SIoT social graph was included
observed, in relation to the size of the graph (number of
vehicles/post of ices 20 CLOs 25 CLOs 30 CLOs 40 CLOs
CLOs). The initialization process entails the instantiation
of the digital twins of logistics objects, along with their 2 CLOs 4.9s 12.4 s 28.7s 95.1s
relevant data structures [28]. Fig. 3 demonstrates how 3 CLOs 9.9s 26.3s 43.4s 168.3s
this process scales in time increasing the number of CLOs 4 CLOs 18.1s 38.2s 78.5s 258.2s
forming the social graph.
5 CLOs 27.5s 52.6s 127.2s 378.4s
The SIoT graph receives update requests every time the
6 CLOs 40.7s 117.4s 228.4s 592.2s
state of CLOs changes. These updates require recalcu‑
lating the relationships between the CLOs. In this ex‑ 7 CLOs 52.8s 172.1s 415.6s 865.4s
periment, the time required for a CLO state update and 8 CLOs 74.3s 230.6s 720.1s 1923.3s
the creation of social relationships between a CLO and
N friends CLOs is observed. Fig. 4 depicts the results
and demonstrates the time necessary to modify the data

76 © International Telecommunication Union, 2021