Uploaded September 2026 | Updated November 2026, 1 day ago
Datacenters are breaking the server apart. Resource disaggregation pools compute, memory, and storage as independent, network-attached resources that can scale on their own, offering elasticity and improved utilization. But breaking the server also breaks traditional assumptions about fault tolerance, latency, and coherence.
In this talk, Aishwarya presents the challenges they are tackling across the disaggregation spectrum, from disaggregated storage to RDMA-based memory to emerging CXL fabrics, in which multiple hosts can share the same physical memory. She presents two systems they have built: SplitFT, which enables low-latency fault tolerance for applications built atop disaggregated storage, and MEGALON, which enables efficient sharing over partly coherent CXL memory.
00:36 Introduction
00:50 Monolithic Servers vs Disaggregation
03:00 Disaggregation Evolution
07:06 Aishwarya's Group's Work
08:43 Compute Express Link (CXL)
09:37 CXL 3.0: Sharing Memory Across Multiple Hosts
10:50 The Promise of CXL & The Partly Coherent CXL Model
17:23 How to Coherently Share Data?
24:19 Introducing Megalon
28:49 How does Software Coherence work with Split Sharing?
32: 15 Key Challenges and the answer: a Shared Log
40:20 Megalon's Solution to CXL's High Latency
41:12 How is Megalon';s Coherence Mechanism Abstracted?
43:47 Megalon Summary
45:14 General Takeaways
Talk from Systems Distributed '26: systemsdistributed.com
linkedin.com/in/aishwaryaganesan
Datacenters are breaking the server apart. Resource disaggregation pools compute, memory, and storage as independent, network-attached resources that can scale on their own, offering elasticity and improved utilization. But breaking the server also breaks traditional assumptions about fault tolerance, latency, and coherence.
In this talk, Aishwarya presents the challenges they are tackling across the disaggregation spectrum, from disaggregated storage to RDMA-based memory to emerging CXL fabrics, in which multiple hosts can share the same physical memory. She presents two systems they have built: SplitFT, which enables low-latency fault tolerance for applications built atop disaggregated storage, and MEGALON, which enables efficient sharing over partly coherent CXL memory.
00:36 Introduction
00:50 Monolithic Servers vs Disaggregation
03:00 Disaggregation Evolution
07:06 Aishwarya's Group's Work
08:43 Compute Express Link (CXL)
09:37 CXL 3.0: Sharing Memory Across Multiple Hosts
10:50 The Promise of CXL & The Partly Coherent CXL Model
17:23 How to Coherently Share Data?
24:19 Introducing Megalon
28:49 How does Software Coherence work with Split Sharing?
32: 15 Key Challenges and the answer: a Shared Log
40:20 Megalon's Solution to CXL's High Latency
41:12 How is Megalon';s Coherence Mechanism Abstracted?
43:47 Megalon Summary
45:14 General Takeaways
Talk from Systems Distributed '26: systemsdistributed.com
linkedin.com/in/aishwaryaganesan










