arXiv subjects

Hong-Yeon Kim

Publications and source records attributed to Hong-Yeon Kim.

4 recordsLinked to original sources

arXiv2026

Bridging LLM Serving and CXL-SSDs with Chunk-Aware KV Cache Management

NAND-backed storage offers the capacity needed to scale LLM prefix caching, but its block I/O path incurs CPU cache contention and host-DRAM staging in addition to NAND latency. Our characterization shows that these interface costs persist even with DRAM as the storage medium, motivating CXL-SSDs for byte-addressable access to NAND-backed capacity. Surprisingly, however, a stock CXL-SSD remains about 3$\times$ slower than local DRAM and no faster than an NVMe SSD, while generic prefetching provides little benefit. We present LM-CXD, a CXL-SSD specialized for LLM prefix caching. LM-CXD bridges the semantic gap between the serving engine, which knows which KV chunks will be consumed, and the device, which controls their placement and movement. It makes KV chunks device-visible I/O units, exposes NAND-to-DRAM progress to the serving engine, and uses device DRAM as a GPU-accessible buffer. LM-CXD further coordinates request scheduling with windowed prefetching and pipelines layerwise KV movement with GPU computation to hide NAND latency under limited device DRAM. Across five LLM models, LM-CXD reduces average TTFT over a stock CXL-SSD by up to 2.6$\times$ with compute asynchronous prefetching and 4.03$\times$ with layerwise prefetching, achieving TTFT within 1.5$\times$ of local DRAM on average.

cs.AR↗

arXiv2025

Shared Disk KV Cache Management for Efficient Multi-Instance Inference in RAG-Powered LLMs

Recent large language models (LLMs) face increasing inference latency as input context length and model size continue to grow. In particular, the retrieval-augmented generation (RAG) technique, which enhances LLM responses by incorporating external knowledge, exacerbates this issue by significantly increasing the number of input tokens. This expansion in token length leads to a substantial rise in computational overhead, particularly during the prefill stage, resulting in prolonged time-to-first-token (TTFT). To address this issue, this paper proposes a method to reduce TTFT by leveraging a disk-based key-value (KV) cache to lessen the computational burden during the prefill stage. We also introduce a disk-based shared KV cache management system, called Shared RAG-DCache, for multi-instance LLM RAG service environments. This system, together with an optimal system configuration, improves both throughput and latency under given resource constraints. Shared RAG-DCache exploits the locality of documents related to user queries in RAG, as well as the queueing delay in LLM inference services. It proactively generates and stores disk KV caches for query-related documents and shares them across multiple LLM instances to enhance inference performance. In experiments on a single host equipped with 2 GPUs and 1 CPU, Shared RAG-DCache achieved a 15~71% increase in throughput and up to a 12~65% reduction in latency, depending on the resource configuration.

cs.AI↗

arXiv2025

Cost-Efficient LLM Serving in the Cloud: VM Selection with KV Cache Offloading

LLM inference is essential for applications like text summarization, translation, and data analysis, but the high cost of GPU instances from Cloud Service Providers (CSPs) like AWS is a major burden. This paper proposes InferSave, a cost-efficient VM selection framework for cloud based LLM inference. InferSave optimizes KV cache offloading based on Service Level Objectives (SLOs) and workload charac teristics, estimating GPU memory needs, and recommending cost-effective VM instances. Additionally, the Compute Time Calibration Function (CTCF) improves instance selection accuracy by adjusting for discrepancies between theoretical and actual GPU performance. Experiments on AWS GPU instances show that selecting lower-cost instances without KV cache offloading improves cost efficiency by up to 73.7% for online workloads, while KV cache offloading saves up to 20.19% for offline workloads.

cs.LG↗

arXiv2024

DeepVM: Integrating Spot and On-Demand VMs for Cost-Efficient Deep Learning Clusters in the Cloud

Distributed Deep Learning (DDL), as a paradigm, dictates the use of GPU-based clusters as the optimal infrastructure for training large-scale Deep Neural Networks (DNNs). However, the high cost of such resources makes them inaccessible to many users. Public cloud services, particularly Spot Virtual Machines (VMs), offer a cost-effective alternative, but their unpredictable availability poses a significant challenge to the crucial checkpointing process in DDL. To address this, we introduce DeepVM, a novel solution that recommends cost-effective cluster configurations by intelligently balancing the use of Spot and On-Demand VMs. DeepVM leverages a four-stage process that analyzes instance performance using the FLOPP (FLoating-point Operations Per Price) metric, performs architecture-level analysis with linear programming, and identifies the optimal configuration for the user-specific needs. Extensive simulations and real-world deployments in the AWS environment demonstrate that DeepVM consistently outperforms other policies, reducing training costs and overall makespan. By enabling cost-effective checkpointing with Spot VMs, DeepVM opens up DDL to a wider range of users and facilitates a more efficient training of complex DNNs.

cs.DC↗