军事知识服务平台
面向大模型推理的存算协同理论综述
A Survey on Compute-Storage Co-Optimization Theory for Large Model Inference
摘要 大模型推理在长上下文、高并发与多轮复用条件下,受算力、存储容量与跨层数据移动共同约束.量化、缓存、卸载与阶段解耦常组合部署,但压缩率、命中率、显存占用下降等局部指标改善,并不等价于端到端关键路径缩短.从权重、键值(KV)缓存与激活的生命周期出发,以关键路径净收益为判据,系统梳理以算换存、以存换算与存算路径3类方法,并归纳其适用条件、附加开销、质量约束与失效边界,为面向负载的存算协同部署提供可复用的分析视角.认为未来应从生命周期控制、压缩表征、跨引擎接口与真实负载评测4个方面开展存算协同相关研究.
Abstract:
Under long context,high concurrency,and multi-turn reuse,large model inference is constrained by computation,storage capacity,and cross-tier data movement.Quantization,caching,offloading,and stage disaggregation are often combined,but improvements in com-pression ratio,hit rate,or memory footprint do not imply a shorter end-to-end critical path.Starting from the lifetimes of weights,key-value(KV)caches,and activations,and using critical-path net benefit as the evaluation criterion,this paper systematically reviews three categories of methods:computation for storage,storage for computation,and storage-computation pathways.It further summarizes their applicability,overheads,quality constraints,and failure modes,and provides a reusable analytical lens for workload-oriented compute-storage co-optimization.This paper holds that future research related to compute-storage co-optimization should be conducted from four aspects:lifecycle control,compressed representation,cross-engine interfaces,and real-workload evaluation.
刊 名 中兴通讯技术 ISTICPKU
年,卷(期): 2026, 32(4)
机标分类号: F270 F830 TP301.6
在线出版日期: 2026年10月21日
基金项目: 2023WNLOKF004:华中科技大学武汉光电国家实验室开放课题计划项目,