|
Work item:
|
Q.MP-GenAI
|
|
Subject/title:
|
Monitoring parameters of network supporting generative artificial intelligence services
|
|
Status:
|
Under study
|
|
Approval process:
|
AAP
|
|
Type of work item:
|
Recommendation
|
|
Version:
|
New
|
|
Equivalent number:
|
-
|
|
Timing:
|
Q1-2028 (Medium priority)
|
|
Liaison:
|
ITU-T SG12, SG13, SG21, IETF RTGWG
|
|
Supporting members:
|
China Unicom, China Telecom, Zhejiang Lab
|
|
Summary:
|
The rapid proliferation of Generative AI (GenAI) introduces unique traffic patterns that challenge conventional network infrastructures. To maintain quality of experience (QoE), optimize costly compute/network resources, and enable granular traffic management, the industry demands advanced monitoring solutions.
Existing technologies (e.g., NetFlow) and standards—designed for traditional web traffic—fail to effectively identify or process GenAI-specific patterns (e.g., AllReduce). This gap necessitates new monitoring capabilities to:
– Drive the evolution of technical standards
– Establish an AI-Ops closed-loop framework
– Support efficient troubleshooting, precision scheduling, and security compliance
A robust GenAI monitoring system is foundational for stable, secure, and high-performance operations. Standardized metrics and parameters will facilitate their consumption by management planes, OSS/NMS, and cross-domain orchestrators (spanning cloud, edge, and core networks).
The metrics defined in this Recommendation are intended to support:
– operational monitoring of GenAI-related traffic and service flows;
– fault detection, localization, and recovery processes; and
performance supervision to ensure efficient resource utilization and the quality of service (QoS) for GenAI applications.
|
|
Comment:
|
-
|
|
Reference(s):
|
|
|
Historic references:
|
|
Contact(s):
|
|
| ITU-T A.5 justification(s): |
|
|
|
|
First registration in the WP:
2026-03-19 14:36:55
|
|
Last update:
2026-04-02 15:07:44
|
|
