Information: Join today and make your research activities more affordable! Technical workshop participation fees and annual registration fees are available at member rates.
Notice: [Important] Announcement of Changes to Registration Fee Payment and Manuscript Upload Procedures for IEICE Technical Meetings
IEICE Technical Committee Submission System
Conference Paper's Information		 Online Proceedings
[Sign in]
Tech. Rep. Archives
 Go Top Page Go Previous   [Japanese] / [English] 

Paper Abstract and Keywords
Presentation 2026-03-04 13:50
Improving Resource Efficiency of Large Language Model Inference via Distributed KV Cache Sharing over All-Photonics Network
Taichi Kondo, Shota Ono, Akihiro Nakao (The Univ. of Tokyo) NS2025-243
Abstract (in Japanese) (See Japanese page) 
(in English) In services powered by large language models, the Key–Value (KV) cache maintained during inference consumes a large amount of memory, placing significant pressure on computational resources and limiting the number of users that can be served concurrently. The objective of this study is to mitigate GPU memory constraints caused by the KV cache and to realize an inference infrastructure capable of accommodating many users while maintaining response quality. To achieve this goal, we propose a distributed KV cache sharing mechanism over the All-Photonics Network (APN). Experimental results show that dynamic KV cache migration increases the number
of supported users by 1.38 × and improves GPU memory utilization by 1.32 × under load imbalance. Furthermore, compared with regenerating the KV cache, the proposed method reduces the Time-To-First-Token (TTFT) by 53.9%, thereby contributing to improved service continuity under limited computational resources.
Keyword (in Japanese) (See Japanese page) 
(in English) LLM / KV Cache / All-Photonics Network / / / / /  
Reference Info. IEICE Tech. Rep., vol. 125, no. 385, NS2025-243, pp. 131-136, March 2026.
Paper # NS2025-243 
Date of Issue 2026-02-25 (NS) 
ISSN Online edition: ISSN 2432-6380
Copyright
and
reproduction		 All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034)
Download PDF NS2025-243

Conference Information
Committee IN NS  
Conference Date 2026-03-04 - 2026-03-06 
Place (in Japanese) (See Japanese page) 
Place (in English) Okinawa-Ken Shichoson Jichi Kaikan 
Topics (in Japanese) (See Japanese page) 
Topics (in English) General 
Paper Information
Registration To NS 
Conference Code 2026-03-IN-NS 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Improving Resource Efficiency of Large Language Model Inference via Distributed KV Cache Sharing over All-Photonics Network 
Sub Title (in English)  
Keyword(1) LLM  
Keyword(2) KV Cache  
Keyword(3) All-Photonics Network  
Keyword(4)  
Keyword(5)  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Taichi Kondo  
1st Author's Affiliation The University of Tokyo (The Univ. of Tokyo)
2nd Author's Name Shota Ono  
2nd Author's Affiliation The University of Tokyo (The Univ. of Tokyo)
3rd Author's Name Akihiro Nakao  
3rd Author's Affiliation The University of Tokyo (The Univ. of Tokyo)
4th Author's Name  
4th Author's Affiliation ()
5th Author's Name  
5th Author's Affiliation ()
6th Author's Name  
6th Author's Affiliation ()
7th Author's Name  
7th Author's Affiliation ()
8th Author's Name  
8th Author's Affiliation ()
9th Author's Name  
9th Author's Affiliation ()
10th Author's Name  
10th Author's Affiliation ()
11th Author's Name  
11th Author's Affiliation ()
12th Author's Name  
12th Author's Affiliation ()
13th Author's Name  
13th Author's Affiliation ()
14th Author's Name  
14th Author's Affiliation ()
15th Author's Name  
15th Author's Affiliation ()
16th Author's Name  
16th Author's Affiliation ()
17th Author's Name  
17th Author's Affiliation ()
18th Author's Name  
18th Author's Affiliation ()
19th Author's Name  
19th Author's Affiliation ()
20th Author's Name  
20th Author's Affiliation ()
21st Author's Name  
21st Author's Affiliation ()
22nd Author's Name  
22nd Author's Affiliation ()
23rd Author's Name  
23rd Author's Affiliation ()
24th Author's Name  
24th Author's Affiliation ()
25th Author's Name  
25th Author's Affiliation ()
26th Author's Name / /
26th Author's Affiliation ()
()
27th Author's Name / /
27th Author's Affiliation ()
()
28th Author's Name / /
28th Author's Affiliation ()
()
29th Author's Name / /
29th Author's Affiliation ()
()
30th Author's Name / /
30th Author's Affiliation ()
()
31st Author's Name / /
31st Author's Affiliation ()
()
32nd Author's Name / /
32nd Author's Affiliation ()
()
33rd Author's Name / /
33rd Author's Affiliation ()
()
34th Author's Name / /
34th Author's Affiliation ()
()
35th Author's Name / /
35th Author's Affiliation ()
()
36th Author's Name / /
36th Author's Affiliation ()
()
Speaker Author-1 
Date Time 2026-03-04 13:50:00 
Presentation Time 25 minutes 
Registration for NS 
Paper # NS2025-243 
Volume (vol) vol.125 
Number (no) no.385 
Page pp.131-136 
#Pages 6 
Date of Issue 2026-02-25 (NS) 

[Return to Top Page]

[Return to IEICE Web Page]
The Institute of Electronics, Information and Communication Engineers (IEICE), Japan