Dichen Li·Oct 12Paper Summary: Mooncake: A KVCache-centric Disaggregated Architecture for LLM ServingThis paper https://arxiv.org/abs/2407.00079 describes an LLM serving architecture built in Moonshot AI.
InITNEXTbyDichen Li·Aug 14Kernel Context Switch Costs And Spin Locks In Perf Profile: A Practical StudySpoiler: On a workload with heavy semaphore wait/post, I observed heavy CPU consumption in perf profiling result in finish_task_switch /…
Dichen Li·Mar 28War story: hunting down a slow mutex. Part II: spinlockIn my previous blog post, I discussed my experience looking into CPU scheduling using perf tool to troubleshoot the mystery of a slow mutex…
Dichen Li·Mar 28War story: hunting down a slow mutex lock. Part I: CPU preemptionAs we are progressing into the Operating Systems: Three Easy Pieces Book Club (organized by the great Phil Eaton), chapter 10 discussed…
Dichen Li·Jan 20My notes taking app story, Part 2: the freedom, and the AI assistanceOne year ago, I wrote about my work notes app story…
Dichen Li·Dec 23, 2025Paper Summary: Theseus: A Distributed and Scalable GPU-Accelerated Query Processing Platform…https://arxiv.org/abs/2508.05029A response icon1A response icon1
Dichen Li·Dec 18, 2025Quick Paper Summary: What Modern NVMe Storage Can Do, And How To Exploit It: High-Performance I/O…https://vldb.org/pvldb/vol16/p2090-haas.pdf
Dichen Li·Dec 16, 2025Understanding Data Storage and Ingestion for Large-Scale Deep Recommendation Model Training: My…This 2022 paper from Meta dive deep into their deep learning training infrastructure, especially storage and data ingestion and…
Dichen Li·Nov 2, 2025Paper summary: Unlocking the Potential of CXL for Disaggregated Memory in Cloud-Native Databaseshttps://dl.acm.org/doi/pdf/10.1145/3722212.3724460
Dichen Li·Oct 31, 2025Quick paper reading note: F3: The Open-Source Data File Format for the Futurehttps://dl.acm.org/doi/pdf/10.1145/3749163A response icon1A response icon1