Applications in modern cloud datacenters are deployed in resource containers to isolate them from each other. Memory stranding is a pervasive problem in such containerized datacenters, where many memory-intensive applications grind to a halt even when free memory exists in other machines. This leads to low utilization, memory fragmentation, and overall increased cost. Memory disaggregation over ultra-fast networks can pool together such stranded memory in theory, but making it practical faces novel systems design, algorithmic, and integration challenges. They include bridging the still-sizable latency gap between local memory access vs. remote memory access; transparently addressing network-wide fault-tolerance, load imbalance, and performance isolation issues; scalability; and enabling support for heterogeneous software and hardware technologies.

The overarching research objective of this project is to realize a Unified Disaggregated Memory (UDM) abstraction over ultra-fast networks to expose stranded memory across the datacenter as a pool of available memory to out-of-memory containers in a fast, resilient, and scalable manner without any changes to the applications. By designing a comprehensive solution to address host-level, network-level, and end-to-end aspects of the aforementioned challenges, this research aims to make memory disaggregation practical. Specifically, by leveraging the unique characteristics of memory-intensive workloads, ultra-low-latency networks, and multi-tenancy in modern datacenters, this proposal will (i) design a low-latency host networking stack; (ii) enable performance isolation throughout the network; (iii) provide resilience to network-wide uncertainties such as failures and load imbalance; and (iv) incorporate support for heterogeneous memory, networking technologies, and resource management software.

People

• Mosharaf Chowdhury (PI)
• Dr. Juncheng Gu → ByteDance
• Dr. Jie You → Meta
• Dr. Hasan Al Maruf → AMD
• Yiwen Zhang

Publications

1. Memory Disaggregation: Advances and Open Challenges, H. A. Maruf, M. Chowdhury, arXiv:2305.03943
2. TPP: Transparent Page Placement for CXL-Enabled Tiered Memory, H. A. Maruf, H. Wang, A. Dhanotia, J. Weiner, N. Agarwal, P. Bhattacharya, C. Petersen, M. Chowdhury, S. Kanaujia, P. Chauhan, ACM ASPLOS, 2023
3. Aequitas: Admission Control for Performance-Critical RPCs in Datacenters, Y. Zhang, G. Kumar, N. Dukkipati, X. Wu, P. Jha, M. Chowdhury, A. Vahdat, ACM SIGCOMM, 2022
4. TPP: Transparent page placement for CXL-Enabled tiered memory, H. A. Maruf, H. Wang, A. Dhanotia, J. Weiner, N. Agarwal, P. Bhattacharya, C. Petersen, M. Chowdhury, S. Kanaujia, P. Chauhan, arXiv:2206.02878
5. Justitia: Software Multi-Tenancy in Hardware Kernel-Bypass Networks, Y. Zhang, Y. Tan, B. Stephens, M. Chowdhury, USENIX NSDI, 2022
6. Hydra: Resilient and Highly Available Remote Memory, Y. Lee*, H. A. Maruf*, A. Cidon, M. Chowdhury, K. G. Shin, USENIX FAST, 2022 (*Equal contribution)
7. Memtrade: A Disaggregated-Memory Marketplace for Public Clouds, H. A. Maruf, Y. Zhong, H. Wang, M. Chowdhury, A. Cidon, C. Waldspurger, arXiv:2108.06893
8. Programmable Packet Scheduling with a Single Queue, Z. Yu, C. Hu, J. Wu, X. Sun, V. Braverman, M. Chowdhury, Z. Liu, X. Jin, ACM SIGCOMM, 2021
9. Ship Compute or Ship Data? Why Not Both?, J. You, J. Wu, X. Jin, M. Chowdhury, USENIX NSDI, 2021
10. Mitigating the Performance-Efficiency Tradeoff in Resilient Memory Disaggregation, Y. Lee, H. A. Maruf, M. Chowdhury, A. Cidon, K. G. Shin, arXiv:1910.09727
11. NetLock: Fast, Centralized Lock Management Using Programmable Switches, Z. Yu, Y. Zhang, V. Braverman, M. Chowdhury, X. Jin, ACM SIGCOMM, 2020
12. Effectively Prefetching Remote Memory with Leap, H. A. Maruf, M. Chowdhury, USENIX ATC, 2020 [hi]Best Paper Award[/hi]

Software

All software artifacts developed as part of this project are released as open-source with permissive licenses and can be found at https://github.com/SymbioticLab.

1. Aequitas source code
2. Justitia source code
3. Memtrade source code
4. AIFO source code
5. Kayak source code
6. Hydra source code
7. NetLock source code
8. Leap source code

Workshop

• NSF Workshop on Next-Gen Cloud Research Infrastructure

Outreach

K-12 students and educators can get involved in this project through the following resources.

• Center for Engineering Diversity & Outreach (CEDO)
• The Data Science Summer Camp for High School Students
• Research Education and Activities for Classroom Teachers (REACT)

Media

1. Meta Platform Hacks CXL Memory Tier into Linux (nextplatform.com)
2. “Hiding” network latency for fast memory in data centers (umich.edu)
3. Chowdhury wins NSF CAREER award for making memory cheaper, more efficient in big data centers (umich.edu)
4. Decentralized Memory Disaggregation Over Low-Latency Networks | USENIX
5. Breakthrough for large scale computing: ‘Memory disaggregation’ made practical | University of Michigan News (umich.edu)
6. Clever RDMA Technique Delivers Distributed Memory Pooling (nextplatform.com)
7. Clever RDMA Technique Delivers Distributed Memory Pooling | Hacker News (ycombinator.com)
8. University of Michigan Demonstrates Novel Memory Disaggregation Technology | TOP500

Support

This project is supported by a CAREER award from the National Science Foundation (CNS-1845853).