Minghao Xie

I am currently seeking job opportunities in the fields of computer networks, systems, and storage technologies.

My name is Minghao Xie, and I am a Computer Engineering Ph.D. candidate at the Baskin School of Engineering, UC Santa Cruz. I hold a B.E. degree in Computer Science from Sichuan University, China.

I am advised by Prof. Chen Qian and Prof. Heiner Litz. My research interests lie in computer networks and systems, with a current focus on flash storage disaggregation within data centers. This work is part of the Center for Research in Systems and Storage (CRSS) and the Hardware Systems Collective (HSC) at UCSC. I am dedicated to building scalable, high-performance, and reliable systems.

I am passionate about research and coding. I am an active language learner, fluent in both Chinese and English, and currently learning Japanese. In my free time, I enjoy spending time with my two cats, Toby and Haruka, who often appear in my recent Instagram posts.

Teaching

CSE150 (TA): Introduction to Computer Networks, 24Spring
CSE120 (TA): Computer Architecture, 24Win, 23Fall, 22Spring, 21Fall, 21Win, 20Win
CMPE110 (TA): Computer Architecture, 19Win

Projects

Thesis: Predictable Cloud Storage System [GitHub], 2019 - 2023
Advisors: Heiner Litz, Chen Qian | Funding: CRSS, NSF

Analyzed the performance unpredictability in ephemeral storage (for serverless computing) and engineered a profile-based system, enhancing goodput predictability by up to 3x.
Created a high-performance, μs-level async serverless client library using Cython, and developed cloud-native applications (e.g. ML and microservices) with it on AWS.
Designed and implemented a novel flow-based scheduler using Ray, resulting in 37% TCO savings without compromising SLO guarantee on IOPS nor tail latency.

Trading Density for Performance in DNA-based Archival Storage Systems, 23Spring
Advisor: Ethan Miller, CSE 290S: Advanced Topics in Computer Systems

Conducted research on the trade-offs between storage density and performance in DNA-based archival storage systems
Designed and evaluated novel approaches to optimize performance without significantly compromising storage density

AIRD: Your Personal AI RSS Daily [GitHub], 2024
Personal Project

Developed an AI-powered RSS assistant in Python for filtering and summarizing articles, featuring advanced deduplication and data privacy.
Configured customizable, multilingual support with plans for enhanced link reading and automated tagging.

Transport Optimization for Storage Disaggregation, June -- Sep 2022
Company: Meta Platform, Mentor: Minkyu Jeong, Position: Research Engineering Intern at CEA

Collaborated with the storage team to identify transport alternatives to Thrift for internal storage workloads, focusing on enhancing performance and scalability.
Achieved a 2.6x increase in throughput and a 63% reduction in latency by optimizing data transport mechanisms and implementing efficient algorithms in C++.
Removed bottlenecks through advanced profiling and tuning techniques, leading to linear throughput scalability and proportional cost efficiency improvements.

Disaggregated Optane SSD for Embedding Table, July -- Sep 2021
Company: Facebook, Mentor: Niket Agarwal, Position: Research Engineering Intern at CEA

Worked closely with the ML team to analyze and address scaling bottlenecks and disaggregation needs of internal PyTorch-based SparseNN workloads.
Developed a fast userspace NBD driver in C++, achieving a speedup of 1.7x to 6.7x, demonstrating proficiency in low-level systems programming and perf. optimization.
Enhanced support for large-scale ML workloads by contributing to an internal library, and published a comprehensive project note to document its impact.

Exploring the New SPDK Threading Model and its Scheduling Possibility, 20Winter
Advisor: Ethan Miller, CSE 231: Advanced Operating Systems

Conducted code analysis and performance evaluations on the new SPDK threading library
Designed and partially implemented a cooperative IO-thread scheduler with new SPDK APIs

Supporting Scalable Fault-Tolerant Distributed Message Queue with Persistent Storage [Github], 19Fall
Advisor: Faisal Nawab, CSE 214: Principles of Database Systems

Designed a distributed message queue framework with fault tolerance
Implemented log-based mechanism to resolve the inter-broker failures

A Ultra-low Latency QUIC-stack Acceleration Engine, 19Winter
Advisor: Scott Beamer, CMPE 293: Programmable Hardware Accelerators

Conducted characterization of QUIC protocol and workload analysis
Designed a NetFPGA-based acceleration engine

Towards Single Round-Trip: An Efficient Metadata Offloading Mechanism for Tree-based Oblivious RAM, 19Winter
Advisor: Chen Qian, CMPE 253: Network Security

Designed a way of efficient metadata management based on ring ORAM and Information-theoretical Private-information-retrieval (ITPIR)
Optimized the algorithm with multi-step bit-based accelerations

Evaluating 100Gbps Flash Disaggregation on ARM SoC [GitHub] [Paper], 2018 - 2019
Advisor: Heiner Litz and Chen Qian, Funding: CRSS, Broadcom \& NSF

Built a performant open-sourced storage system with SPDK and DPDK in C.
Optimized the ARM system at the assembly level, reducing TCO by 2.57x.

Publications

Liu, Y., Shi, S., Xie, M., Litz H., \& Qian, C. SMASH: Flexible, Fast, and Resource-efficient Placement and Lookup of Distributed Storage. In conference of the ACM special interest group for the computer systems performance evaluation community (SIGMETRICS ‘23).
Shi, S., Yu, Y., Xie, M., Li X., Li, X., Zhang, Y. \& Qian, C. Concury: A Fast and Light-weight Software Cloud Load Balancer. In conference of the ACM Symposium on Cloud Computing 2020 (SOCC ‘20).
Xie, M., Qian, C., Litz, H. ReFlex4ARM: Supporting 100GbE Flash Storage Disaggregation on ARM SoC. OCP Future Technologies Symposium 2020. Poster session in San Jose, CA (OCP ‘20).
Xie, M., Qian, C., Litz, H. (2019, August 4-6). Evaluating 100Gbps Flash Disaggregation on ARM SoC. Flash Memory Summit 2019. Poster session in Santa Clara, CA.

Last update on 2024-08-03