The CAOS group at CMU bridges computer
architecture and operating systems
and rethinks their synergy for performance,
security, and scalability in AI Datacenters
Datacenter computing is undergoing a transformative shift, driven by the unprecedented growth of AI, data, and user demand. Yet, the hardware and operating system we rely on were designed for an era devoid of AI. Datacenters must contend with challenges spanning the hardware and software stack. At the hardware level, the deployment of terabyte-scale heterogeneous memory systems and the proliferation of AI hardware introduces significant energy and efficiency concerns. On the software side, the growing adoption of compound AI systems is unlocking new opportunities to apply machine learning toward improving system design and operational efficiency; however, it has also highlighted severe inefficiencies in legacy software and hardware stacks, leading to substantial resource underutilization and performance overheads. Meanwhile, the increasing dependence on data-hungry AI models intensifies the challenge of safeguarding sensitive information, as users and organizations expect strict privacy guarantees. Our research at the CAOS group aims to fundamentally re-architect the hardware and software stack to address the demands of modern AI datacenters. We explore new abstractions and mechanisms that improve performance, resource efficiency, scalability, while delivering strong privacy guarantees.
Recent News
-
2025
LithOS accepted at SOSP 2025!
Learn more about our work on an Operating System for Efficient Machine Learning on GPUs.
-
2025
LVM accepted at MICRO 2025!
Learn more about our work on architecting Learned Virtual Memory Translation.
-
2025
Dimitrios received the IEEE CS TCCA Young Computer Architect Award!
• Learn more about the award.
• CMU School of Computer Science news. -
2025
Initial patches of Contiguitas have been upstreamed into Linux 6.15!
-
2025
Joshua was selected for the Allen Newell Award for Excellence in Undergraduate Research!
-
2025
Dimitrios received the Amazon Research Faculty Award!
-
2025
Dimitrios received the Wilton E. Scott Institute for Energy Innovation Faculty Award!
-
2025
EMT accepted at OSDI 2025!
Learn more about our work on an OS framework for new nemory translation architectures.
-
2025
Joshua was selected for the CRA Outstanding Undergraduate Research Honorable Mention!
-
2024
Cinnamon accepted at ASPLOS 2025!
Learn more about our work on a framework for scale-out encrypted AI.
-
2024
Dimitrios received the Intel Rising Star Faculty Award!
-
2024
Dimitrios received the Linux and eBPF Foundation Faculty Award!
-
2024
Eliot and Vasilis join the CAOS group!
Welcome!
-
2024
Kaiyang and Hilbert (Yuang) win the Qualcomm Innovation Fellowship 2024!
• Learn more about the Qualcomm Innovation Fellowship.
• CMU School of Computer Science news. -
2024
Perspective accepted at ISCA 2024!
•Learn more about our work on a principled framework for secure speculation in operating systems.
• CMU CyLab news. -
2024
Contiguitas selected as an IEEE MICRO Top Picks!
Learn more about our work on the pursuit of physical memory contiguity in datacenters.
-
2024
David Rudo was selected for the CRA Outstanding Undergraduate Research Honorable Mention!
-
2023
Hilbert joins the CAOS group!
Welcome!
-
2023
Memento accepted at MICRO 2023!
Learn more about our work on ephemeral memory management in serverless environments.
-
2023
Two papers, Siloz, and XFaaS, accepted at SOSP 2023!
Learn more about our work on:
• DRAM isolation.
• Hyperscale and frugal Function-as-a-Service. -
2023
Contiguitas won the best paper ISCA 2023!
•Learn more about our work on the pursuit of physical memory contiguity in datacenters.
•CMU School of Computer Science news. -
2023
Contiguitas accepted at ISCA 2023!
Learn more about our work on the pursuit of physical memory contiguity in datacenters.
-
2023
ServiceRouter accepted at OSDI 2023!
Learn more about our work on a minimal service mesh for datacenters.
-
2023
Dimitrios received the NSF CAREER Award!
Learn more about our work on rebuilding virtual memory across hardware and operating systems.
-
2023
IOCost selected as an IEEE MICRO Top Picks!
Learn more about our work on block IO control for containers in datacenters.
-
2022
Memory Efficent Hashed Page Tables accepted at HPCA 2023!
Learn more about our work on Memory Efficient Hashed Page Tables.
-
2022
Siddharth and Patrick join the CAOS group!
Welcome!
-
2022
Dimitrios received three Meta research awards to improve datacenter computing systems!
Learn more about the awards:
• Meta Security Research Award.
• Meta Systems Research Award.
• Meta AI Hardware/Software Codesing.
• CMU School of Computer Science news. -
2022
TMO won the best paper award at ASPLOS 2022!
Learn more about our work on transparent memory offloading in datacenters.
-
2022
Three papers, IOCost, TMO, and Nested ECPTs, accepted at ASPLOS 2022!
Learn more about our work on:
• Block IO control for containers in datacenters.
• Transparent memory offloading in datacenters.
• Parallel virtualized memory translation. -
2021
Dimitrios selected as a Meta Academic of the Month!
-
2021
Dimitrios received a Meta Faculty Award!
For his work on architectural and operating system support for lightweight virtualization.
-
2021
Dimitrios received the David J. Kuck Outstanding Ph.D. Thesis Award!
Awarded to a single outstanding doctoral thesis from the computer science department at UIUC each year.
-
2021
RAS accepted at SOSP 2021!
Learn more about our work on continuously optimized region-wide datacenter resource allocation.
-
2021
Dimitrios received the ACM SIGARCH & IEEE CS TCCA Outstanding Dissertation Award!
For "contributions to redesigning the abstractions and interfaces that connect hardware and operating systems".
-
2021
NVOverlay accepted at ISCA 2021!
Learn more about our work on efficient and scalable high-frequency snapshotting to NVM.
-
2021
Kaiyang and Ziqi join the CAOS group!
Welcome!
-
2021
BabelFish selected as an IEEE MICRO Top Picks!
Learn more about our work on fusing address translations for containers.
-
2021
Elastic Cuckoo Page Tables selected as a a MICRO Top Picks Honarable Mention!
Learn more about our work on rethinking virtual memory translation for parallelism.
-
2021
The upstreamed Draco is featured in Linux Security at Phoronix!
-
2021
Jamais Vu accepted at ASPLOS 2021!
Learn more about our work on thwarting microarchitectural replay attacks.
-
2021
Draco has been upstreamed in the Linux Kernel!
Learn more about our work on architectural and operating system support for system call security.
-
2021
CAOS Group Established!
And so it begins...