CV

Education

• M.S. in Computer Technology, Shanghai Jiao Tong University, 2022 ———— Present
• B.S. in Computer Science and Technology, Shanghai Jiao Tong University, 2022

Publications

HPool: High-performance Memory Pool Management through Multi-stream Data Swap

Boosting Data Center Performance via Intelligently Managed Multi-backend Disaggregated Memory

Teaching experience

• CS3328 - Cloud Computing Technology, Shanhai Jiao Tong University, Sept. 2024 ———— Present
  • Teaching Assistant
  • Duties included: Design the course experiments and projects.

Working experience

• Far data lab, remote, April. 2025 ———— Present

  Research Assistant

Projects

• LLM Agent Framework Construction and Related Researches

  Main Contributor, Jul. 2024 ———— Present

  • Researching the system for LLM agents, especially the context (including agent ”memory”, tool results, and external knowledge) cache and position-independent KV cache system.

• Optimization of LLM Inference Framework(Based on vLLM)

  Contributor, Jan. 2024 ———— Jun. 2024

  • Built a prefix tree to cache KV during inference, accelerating multi-round chat, shared knowledge chat (RAG), and multi-user chat, resulting in a 22.8% reduction in inference TTFT compared to vLLM.
  • Disaggregated the prefill and decode stages, distributing them across different GPUs for cross-cluster deployment, which improved overall throughput by 25% compared to vLLM.

• High-bandwidth Multi-stream Parallel Data Swap System in Memory Pool

  Main Contributor, Sept. 2023 ———— Feb. 2024

  • Designed and implemented a service-aware memory offloading mechanism with a multi-queue structure for the memory pool in cloud computing, based on the memory access patterns of different services.
  • Implemented a data swap subsystem with stream-level and path-level parallelism to enhance data offloading performance, achieving a 50% memory reduction with less than 3% time overhead when integrated with the existing far memory system.

• Application-Aware Multi-Path Far Memory Management System

  Main Contributor, Jun. 2022 ———— Apr. 2023

  • Designed and implemented a memory system that leverages the VM kernel isolation mechanism to achieve application transparency and awareness, isolating swap paths between different applications to enable parallel swaps.
  • Designed and implemented an optimization system for backend selection and swap path parameter adjustment, achieving up to 3.9x faster data swap performance and a 5.1x improvement in data center task throughput compared to SOTA systems.

Skills

Programming Language: C/C++, Python, SQL, Latex

Soft Skills: Linux; Linux kernel (memory part); RDMA; Pytorch; ML; LLM; Big data.

Language: English (TOEFL 109: R:29, L:30, S:22, W:28)