ryan@cl:~$ whoami
computer systems · data centers · AI/ML efficiency

Ryan Sana

I find problems and make them disappear, for fun and profit.
CXL & memory systems kernel / QEMU datacenter networking ML systems energy efficiency
Ryan Sana

I build high performance systems for modern data centers, with a focus on AI and ML workloads and their efficiency at scale. I am comfortable moving across problems, but my core experience is in systems software. You can find me at the Computer Laboratory, where I am visiting Andrew Moore at the University of Cambridge.

I spent about two years at Huawei working on Compute Express Link (CXL) across QEMU and the Linux kernel, exploring how memory systems should evolve for modern data centers. More recently I have focused on AI and ML workloads and the systems challenges they introduce — optimizing them at scale across networking, memory, and runtime layers, with a growing emphasis on energy efficiency alongside performance.

My interests center on computer systems for data centers, especially where real systems meet emerging workloads. I enjoy technical discussions and early stage ideas, and I am always open to exchanging thoughts.

§01

Background

I received my PhD from Queen Mary University of London, advised by Gianni Antichi and Brent Stephens (Utah / Google) — see my PhD dissertation. Beyond their technical depth, they each have distinct styles that strongly shaped how I think about systems and research.

Before the PhD I was an intern and later a Research Assistant at NUS, advised by Djordje Jevdjic, working on memory management for data-center applications. I also spent three months at EPFL, studying virtualization to improve the availability of electrical grid controllers, advised by Jean-Yves Le Boudec. Lausanne is beautiful — pay it a visit. I completed my undergraduate and master's studies at IUST, Iran, working with Mohsen Sharifi.

§02

Selected Publications

Enabling Fast Networking in the Public Cloud
A. Sanaee, V. Jabrayilov, I. Marinos, F. Shahinfar, D. Saxena, G. Antichi, K. Kaffes
ACM ASPLOS '26papercode
IPA: Inference Pipeline Adaptation to Achieve High Accuracy and Cost-Efficiency
S. Ghafouri, K. Razavi, M. Salmani, A. Sanaee, T. Botran, L. Wang, J. Doyle, P. Jamshidi
ICPE '25
Scalable and Effective Page-table and TLB Management on NUMA Systems
B. Gao, Q. Kang, H. Tee, K. Chu, A. Sanaee, D. Jevdjic
USENIX ATC '24papertalk
Backdraft: a Lossless Virtual Switch that Prevents the Slow Receiver Problem
A. Sanaee, F. Shahinfar, G. Antichi, B. Stephens
USENIX NSDI '22papertalkpodcast
Morpheus: Domain-Specific Run Time Optimization for Software Data Planes
S. Miano, A. Sanaee, G. Retravi, G. Antichi
ACM ASPLOS '22papertalkcodepodcast

See all publications on Google Scholar →

§03

Workshops

Reconciling High Accuracy, Cost-Efficiency, and Low Latency of Inference Serving
M. Salmani, S. Ghafouri, A. Sanaee, K. Razavi, M. Mühlhäuser, J. Doyle, P. Jamshidi, M. Sharifi
MLSys Workshop @ EuroSys '23papertalk
§04

Teaching & Mentoring

I enjoy mentoring students and working closely on open-ended projects.

previous mentees
§05

Service

§06

Some Thoughts

Occasional notes, opinions, and unfinished ideas.