DUAL-BLADE: Dual-Path NVMe-Direct KV-Cache Offloading for Edge LLM Inference

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jeong, Bodon, Byun, Hongsu, Kim, Youngjae, Yu, Weikuan, Lee, Kyungkeun, Yang, Jihoon, Park, Sungyong
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915967529910272
author Jeong, Bodon
Byun, Hongsu
Kim, Youngjae
Yu, Weikuan
Lee, Kyungkeun
Yang, Jihoon
Park, Sungyong
author_facet Jeong, Bodon
Byun, Hongsu
Kim, Youngjae
Yu, Weikuan
Lee, Kyungkeun
Yang, Jihoon
Park, Sungyong
contents The increasing deployment of Large Language Model (LLM) inference on edge AI systems demands efficient execution under tight memory budgets. A key challenge arises from Key-Value (KV) caches, which often exceed available device memory. Although NVMe-based offloading offers scalable capacity, existing file-based designs rely heavily on the kernel page cache, leading to cache thrashing, unpredictable latency, and high software overhead under memory pressure. We present DUAL-BLADE, a dual-path KV residency framework that dynamically assigns KV tensors to either a page-cache path or an NVMe-direct path based on runtime memory availability. The NVMe-direct path bypasses the filesystem by mapping KV tensors to contiguous logical block address (LBA) regions, enabling low-overhead direct storage access. DUAL-BLADE further incorporates adaptive pipeline parallelism to overlap storage I/O with GPU DMA, improving inference throughput. Our evaluation shows that DUAL-BLADE substantially mitigates I/O bottlenecks, reducing prefill and decode latency by up to 33.1% and 42.4%, respectively, while improving SSD utilization by 2.2x across diverse memory budgets.
format Preprint
id arxiv_https___arxiv_org_abs_2604_26557
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle DUAL-BLADE: Dual-Path NVMe-Direct KV-Cache Offloading for Edge LLM Inference
Jeong, Bodon
Byun, Hongsu
Kim, Youngjae
Yu, Weikuan
Lee, Kyungkeun
Yang, Jihoon
Park, Sungyong
Distributed, Parallel, and Cluster Computing
Artificial Intelligence
Performance
The increasing deployment of Large Language Model (LLM) inference on edge AI systems demands efficient execution under tight memory budgets. A key challenge arises from Key-Value (KV) caches, which often exceed available device memory. Although NVMe-based offloading offers scalable capacity, existing file-based designs rely heavily on the kernel page cache, leading to cache thrashing, unpredictable latency, and high software overhead under memory pressure. We present DUAL-BLADE, a dual-path KV residency framework that dynamically assigns KV tensors to either a page-cache path or an NVMe-direct path based on runtime memory availability. The NVMe-direct path bypasses the filesystem by mapping KV tensors to contiguous logical block address (LBA) regions, enabling low-overhead direct storage access. DUAL-BLADE further incorporates adaptive pipeline parallelism to overlap storage I/O with GPU DMA, improving inference throughput. Our evaluation shows that DUAL-BLADE substantially mitigates I/O bottlenecks, reducing prefill and decode latency by up to 33.1% and 42.4%, respectively, while improving SSD utilization by 2.2x across diverse memory budgets.
title DUAL-BLADE: Dual-Path NVMe-Direct KV-Cache Offloading for Edge LLM Inference
topic Distributed, Parallel, and Cluster Computing
Artificial Intelligence
Performance
url https://arxiv.org/abs/2604.26557