IEEE Consumer Electronics Magazine - May/June 2023 - 103

FIGURE 5. Average service discovery time. (a) In the
four cases. (b) Under different number of users per EC.
acts as a centralized controller to manage all service
information. A service can be discovered in
the following four cases.
Initialization (case 0): No user in the distributed
cloud has requested the service yet.
At CC (case 1): No user in the current RC has
requested the service yet, but a user from
another RC has already requested the service.
At RC (case 2): No user in the current EC has
requested the service yet, but a user from
another EC managed by the same RC has
requested it.
At EC (case 3): The service has already been
requested by another user in the current EC.
Figure 5(a) illustrates the average discovery
time in the four cases, each of which is tested
with 100 runs of sending a user's service request.
In case 0, the hierarchical architecture takes
2.73 seconds to communicate from the EC to the
SP while the centralized architecture takes about
2.33 seconds. Besides, both take about 7 seconds
to initialize the container to run the service. In
case 1, the service discovery time of the hierarchical
architecture is 2.1 seconds, which is longer
than that of the centralized architecture with
1.7 seconds. The results in these two cases can be
explained by the fact that in the hierarchical architecture,
the discovery process in the RC causes a
little delay, while in the centralized architecture,
the EC just forward the request to the centralized
controller at the CC. However, in case 2 where the
service has been requested at other EC in the
same region, the hierarchical architecture can
save time up to 0.59 seconds compared to the centralized
architecture. If the service instance is running
in the current EC (i.e., case 3), both achieve
the sametimeof0.11 seconds.
Next, we increase the number of users in
each EC and evaluate the performance of the
May/June 2023
two architecture in terms of average service discovery
time, which is calculated based on the
measured service discovery time of the four
cases and their probability of occurrence.
Figure 5(b) illustrates that the average service
discovery time decreases further as the number
of users increases. It is because the probability
of occurrence of case 3 increases as the number
of users increases. It means that the average service
discovery time will converge to the case
where the user requests are served by their connected
ECs. Moreover, the result shows that the
hierarchical outperforms the centralized in
terms of average service discovery time.
OPEN RESEARCH CHALLENGES
Heterogeneity
Heterogeneity is a main challenge to distributed
cloud that arises from not only the heterogeneous
infrastructure resources in different
technology domains of compute, network, and
storage but also the different requirements of
service functions, including cloud-related and
networking-related functions.4 To address the
heterogeneity challenges, abstraction models
should be developed and standardized on both
resource and service levels to support multilevel
virtualization, including virtual infrastructure,
virtual networks, and virtual functions. Research
efforts toward resource and service modeling
for unified abstraction across domains are fairly
recent and more study is needed.
Scalability and Performance
In distributed cloud, various virtual functions
are realized as service components distributed
scatterly across network domains. The adoption of
cloud-native trend poses even more scalability
issues due to the large scale of microservices. On
the other hand, service performance guarantee is
not just about network latency, but also how
quickly and efficiently resources can be made
available for a service as needed, especially as it
scales. To build scalable and performant microservices-based
distributed cloud systems, optimal
workload placement, capacity planning, traffic
management, and resource management across
domains are all open issues to be fully studied.
103
IEEE Consumer Electronics Magazine - May/June 2023

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - May/June 2023
