|
Work item:
|
Y.IMT2020-qos-aimtdc
|
|
Subject/title:
|
QoS assurance requirements and framework for artificial intelligence model training over distributed cloud supported by IMT-2020 networks and beyond
|
|
Status:
|
Under study
|
|
Approval process:
|
AAP
|
|
Type of work item:
|
Recommendation
|
|
Version:
|
New
|
|
Equivalent number:
|
-
|
|
Timing:
|
2027-07 (Medium priority)
|
|
Liaison:
|
ITU-T SG11, IETF CATS
|
|
Supporting members:
|
China Unicom, ZTE Corporation, University of Science and Technology Beijing，Zhejiang Lab
|
|
Summary:
|
Driven by big data and artificial intelligent technologies, artificial intelligence(AI) model training over distributed cloud have rapidly expanded into diverse fields such as scientific research, healthcare, transportation, and multimedia production. AI model training has become a key part of digital transformation, impacting economy and society, and heralding the intelligence era.
AI model training over distributed cloud involves complex scenarios and requirements, which can be generally classified into typical scenarios: the uploading of massive data to the data center, AI training in the data center, collaborative AI training among multiple data centers, and the execution of diverse inference tasks. That imposes QoS assurance requirements on network performance such as bandwidth, throughput, latency, and packet loss rate
In the current network environment, AI model training over distributed cloud encounter challenges during cross-regional and large-scale network deployments, such as network congestion, high data transmission costs, limited bandwidth, low throughput, significant RTT latency, low utilization of network, inefficient model training, unstable real-time inference etc. Therefore, it is necessary to establish consistent QoS assurance requirements and frameworks for artificial intelligence(AI) model training over distributed cloud, providing stable, reliable, and high-performance network services to improve their efficiency and user experience.
This new Recommendation is to specify the QoS considerations for typical scenarios, formulating the essential QoS requirements, and providing the framework to supports the QoS assurance for AI model training over distributed cloud.
|
|
Comment:
|
-
|
|
Reference(s):
|
|
|
Historic references:
|
|
Contact(s):
|
|
| ITU-T A.5 justification(s): |
|
|
|
|
First registration in the WP:
2025-03-19 12:33:50
|
|
Last update:
2025-04-23 18:34:09
|
|
