|
Work item:
|
Q.CPNG-DI
|
|
Subject/title:
|
Architecture and signalling requirements of computing power network gateway supporting distributed inference
|
|
Status:
|
Under study
|
|
Approval process:
|
AAP
|
|
Type of work item:
|
Recommendation
|
|
Version:
|
New
|
|
Equivalent number:
|
-
|
|
Timing:
|
2028-07 (Medium priority)
|
|
Liaison:
|
ITU-T SG13, SG2, IETF RTGWG, BMWG
|
|
Supporting members:
|
China Telecom, CAICT, China Unicom
|
|
Summary:
|
With the rapid development of Large Language Models (LLMs), large-scale inference services are facing increasing challenges in terms of high concurrency, long-context processing, and frequent access to key-value (KV) cache. As a result, distributed inference becomes an important technical scheme to improve inference performance and resource utilization. This document aims to study the architecture and signalling requirements of border network gateway for distributed inference in computing power network, so as to support multi-tenant isolation, inference resource telemetry, KV cache routing and flow control, and improve the flexibility and efficiency of distributed LLM inference deployment.
|
|
Comment:
|
-
|
|
Reference(s):
|
|
|
Historic references:
|
|
Contact(s):
|
|
| ITU-T A.5 justification(s): |
|
|
|
|
First registration in the WP:
2026-08-18 13:52:23
|
|
Last update:
2026-08-18 14:04:40
|
