采用高带宽ReRAM近内存架构的分散式大语言模型（LLM）服务中的MoE专家执行
MoE Expert Execution in Disaggregated LLM Serving with a High-Bandwidth ReRAM Near-Memory Architecture
浏览论文内容
中文总结 AI 辅助
本文针对分散式LLM服务中MoE专家执行的资源效率问题，提出ReRAM近内存架构，通过优化池化、放置与取数策略提升资源占用率，在实测中大幅降低延迟与能耗。
中文摘要 AI 辅助
注意力-前馈网络（FFN）分散化将LLM模块映射至专用资源池，为将混合专家（MoE）权重驻留在高带宽FFN资源池创造了条件。然而，解码服务水平目标（SLO）限制了运行批次规模，而稀疏路由会扩大激活专家的集合，导致权重流量分摊效果差，且路由偏差会造成冷专家资源闲置。因此，FFN资源池需在稀疏集合下提供权重读取带宽密度，并在无全局共享结构的情况下恢复资源占用率。本文提出一种ReRAM近内存架构，将专家权重驻留在高带宽本地读取路径后。该设计将实际模型利用率（MFU）分解为理想MFU和资源占用率，通过有界核心本地多播池、共激活感知放置及负载感知取数恢复资源占用率，并根据诱导需求确定各通信层级的规模。针对Qwen3.5-35B-A3B、Qwen3.5-397B-A17B和GLM-5.2的实测与建模研究显示，4路池化将资源占用率从0.328提升至0.519，在峰值计算量相同的情况下，与H20相比，每token的FFN池延迟降低9.5倍，权重移动能耗降低20倍；H20注意力+ReRAM-FFN系统的解码每token输出时间（TPOT），与同构H20池相比分别降低1.25-4.0倍、2.4-10.3倍和2.5-10.4倍。
英文摘要
Attention-FFN disaggregation maps LLM modules to specialized pools, creating an opening to keep Mixture-of-Experts (MoE) weights resident in a high-bandwidth FFN pool. Decode SLOs, however, cap the run-batch while sparse routing expands the activated-expert union, so weight traffic amortizes poorly and routing skew idles cold-expert resources. The FFN pool must therefore deliver weight-read bandwidth density under sparse unions and recover occupancy under skew without a global sharing fabric. We present a ReRAM near-memory architecture that keeps expert weights resident behind high-bandwidth local reads. The design factors actual MFU into ideal MFU and occupancy, recovers occupancy with bounded core-local multicast pooling, coactivation-aware placement, and load-aware fetch, and sizes each communication level from induced demand. A measured + modeled study on Qwen3.5-35B-A3B, Qwen3.5-397B-A17B, and GLM-5.2 shows that side-4 pooling raises occupancy from 0.328 to 0.519 and, at iso-peak compute, lowers per-token FFN-pool latency by 9.5x versus H20 with 20x lower weight-movement energy; an H20-attention + ReRAM-FFN system reduces decode TPOT by 1.25-4.0x, 2.4-10.3x, and 2.5-10.4x versus a homogeneous H20 pool.
