This paper presents HyMCache, a CXL memory rack for multi-turn LLM serving using cost-efficient CXL-hybrid memory, which combines a small amount of in-device DRAM with large SSD-backed capacity behind a CXL interface to efficiently support TB-scale SSD-backed KV reuse.
Abstract
Long-context, multi-turn, and agentic LLM workloads increasingly reuse previously processed context, making KV-cache reuse essential for reducing redundant computation. However, this reuse shifts the bottleneck to the memory tier that stores and serves reusable KV states at cluster scale. GPU HBM and host DRAM are too costly to scale to TB-scale shared context capacity, motivating remote tiers built from lower-cost, higher-capacity media. This paper presents HyMCache, a CXL memory rack for multi-turn LLM serving. We build the memory rack using cost-efficient CXL-hybrid memory (CXL-HM), which combines a small amount of in-device DRAM with large SSD-backed capacity behind a CXL interface. By exploiting the read-dominant, predictable, and append-only nature of multi-turn KV-cache access, HyMCache rethinks DRAM management within CXL-HM to efficiently support TB-scale SSD-backed KV reuse. It uses request-level prefix prefetching and opportunistic write buffering to stage latency-critical reads in device DRAM, enabling DRAM-scale KV-cache efficiency at SSD-level cost. We evaluate HyMCache on a real CXL-HM prototype under both single-aggregator and PD-disaggregated serving configurations. Under the same DRAM budget, HyMCache outperforms local LMCache by 3.0x in single-node serving and 1.45x in PD-disaggregated serving. Compared with 1 TB distributed-DRAM Mooncake, HyMCache incurs about 30% lower performance but uses 16x less DRAM.
We present a Kubernetes Dynamic Resource Allocation (DRA) driver that makes composable CXL memory a schedulable cluster resource, and evaluate the resulting shared-memory tier for cross-node KV-cache reuse in LLM serving. The driver composes CXL regions on demand, materializes them as DAX devices on each participating host, and injects them into pods under a single Container Device Interface (CDI) name so that pods on different nodes access the same physical region. A shared-memory connector for vLLM/llm-d uses that region as a KV-cache tier with a slot directory embedded inside the shared medium, which eliminates the need for an external metadata service. On a two-node cluster with a 512\,GiB CXL appliance and Qwen2.5-7B-Instruct, cross-node prefix reuse reduces TTFT by 5.5$\times$--36.6$\times$ at an external hit rate of 95.4--99.5\,\%, while node-local tiers (GPU prefix caching, CPU-DRAM offload) fall back to full recompute. The sharing gap, defined as the latency ratio between cross-node and same-node reuse, is 1--4\%, indicating that cross-node reuse incurs little additional latency relative to same-node reuse on our testbed. Both replicas run full engines; the study demonstrates memory disaggregation rather than prefill/decode disaggregation. We report this as a feasibility study rather than a performance evaluation.
Hong-Jiang Fan, Ke-Rao Zhang, David Habinsky et al.· 0 citations
OasisKV is presented, a memory-centric LLM inference system design that alleviates HBM capacity pressure by decoupling full KV-cache storage from HBM during LLM decoding and observes that future important tokens can be predicted accurately in advance using lookahead tokens drafted by speculative decoding (SD).
This work presents the Marvell Photonic Fabric Memory Appliance, a photonic-CXL hybrid architecture replacing electrical switches with a passive fiber shuffle to deliver 32 TB shared memory across 16 hosts via a switch-free full- crossbar topology.
This model reveals one key opportunity: dividing a restore request proportionally between the storage path and the GPU can improve inference performance while still meeting SLOs, and reduces the KV-cache storage stack to a performance model based on per-tier capacity, per-tier and interconnect bandwidth, and GPU arithmetic throughput to identify optimization opportunities for KV-cache management.
DynamoServe is presented, a multi-tenant LLM serving framework that addresses challenges through three key innovations: leveraging stranded GPU memory to offload model weights and KV caches, mitigating resource fragmentation in multi-workload environments, and improving memory locality through coordinated data placement and demand-driven weight migration across GPUs.
Diman Zad Tootaghaj, Khaled Diab, Bob Lantz et al.· Conference on Applications,...· 0 citations
Results show that object-value signals rank what to retain, while persistent responsibility determines which group bears reclamation pressure, which shows that object-value signals rank what to retain, while persistent responsibility determines which group bears reclamation pressure.
Zhi-Yu Wang, Rajkummar Buyya· 1 citation
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
