HyMCache: A KV Cache Framework for Multi-Turn LLM Serving with CXL-Hybrid Memory
2026-07-20 • Distributed, Parallel, and Cluster Computing
Distributed, Parallel, and Cluster Computing
AI summaryⓘ
The authors developed HyMCache, a system to efficiently reuse memory called KV-cache for large language models working with long conversations. They use a special hybrid memory combining fast but small DRAM with slower but larger SSD storage, connected via CXL technology. By smartly preloading and buffering data, HyMCache makes it possible to handle huge amounts of memory with lower costs while keeping performance high. Their tests show HyMCache works much faster than similar local memory systems and uses far less expensive DRAM than big distributed setups with only a small performance drop.
KV-cacheLarge Language ModelsCXL (Compute Express Link)Hybrid MemoryDRAMSSDPrefetchingMulti-turn ServingMemory ManagementDistributed Systems
Authors
Hakbeom Jang, Inho Song, Sam H. Noh, Jongryool Kim
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 KV-cache framework that integrates CXL-hybrid memory (CXL-HM) for multi-turn LLM serving. CXL-HM 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.