Page 112 - ITU Journal Future and evolving technologies Volume 2 (2021), Issue 4 – AI and machine learning solutions in 5G and future networks

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 4

2.7 Standard packaging and rapid promotion
Lin Guo said, ‘In order to meet the needs of rapid
replication and promotion, according to ITU ML
architecture and recommendations, the
requirements in ITU recommendations are
implemented by docker and Kubernetes’. Docker
container is a standard unit of software that
packages up code and all its dependencies so the
Fig. 16 – Increased 11%
application runs quickly and reliably from one
2.6 Real-world scenario computing environment to another K8s, is an open-
source system for automating deployment, scaling,
Qianyin Rao introduced the results into Kaili City,
with 913 chains composed of 3191 network and management of containerized applications.
elements, and an AI optimization algorithm output According to the test, the development time is
of 60 high-load solutions, of which 29 have passed shortened by 70%, data sharing is applied to reduce
the expert review and are qualified for 50% storage, and fault self-healing is realized.
implementation. In order to ensure safety, the Finally, the overall architecture is completed on the
operation is carried out in batches. Each basis of integration with ITU architecture.
optimization does not exceed 50 nodes, and the RESULTS AND DISCUSSION
scope of influence is controlled within 10 links. 3.
After the operation is completed, observe the 3.1. Results
stability of the network for 3 days before
proceeding to the next operation. Up to now, 11 In accordance with the ITU standard Lin Guo
batches of adjustments have been completed, and evaluated the results, the application results
16 high-load links have restored reasonable loads. reached L3 in analysis, decision and demand
mapping, and the data collection reached L2. Since
the newly-built link in the transmission topology
optimization involves physical entity operations
and cannot be fully implemented by the software
system, the action implementation only reaches L1.
But the team believes that after the results continue to
improve, the three parts of data collection, analysis
and decision can be upgraded from the current L2 and
L3 levels to L5. The demand mapping team believes
that it needs to accumulate more expert experience to
Fig. 17 – PTN ring utilization break through to L4. If the results are applied to other
similar network topology optimizations that do not
For these 16 overloaded links, compared with the require physical adjustment, the team believes that L5
traditional solution, this network topology can be achieved.
optimization saves construction costs by 2.6 million
RMB, reduces personnel investment by 17 man-days,
and increases network absorption traffic by 79TB/day.

Fig. 19 – Intelligent level
Fig. 18 – Economic performance At the same time, the team believes that there is still
room for further evolution of this result.

96 © International Telecommunication Union, 2021