Research
An understanding of access patterns to byte-addressable memory is fundamental for the allocation and mapping of memory that is close to CPUs and accelerators or even far away (disaggregated memory) via network technologies such as Computer Express Link (CXL). more
In case of persistent memory technologies with hybrid access modes (byte- and block-addressable), the operating System can determine the optimal mode dynamically. more
Our approach to virtual memory management can allocate, map, and migrate memory areas to the optimal memory class (e.g., volatile or persistent memory and storage ) depending on run-time criteria like capacity, bandwidth, latency, and energy consumption. more
File systems for accelerators used in areas such as AI and HPC are an alternative approach to traditional CPU-based file access. If accelerators have full control over the data path from the storage media to the accelerator's memory, they can reduce latencies, discharge the CPU from management work, and don't waste the valuable main memory bandwidth for the I/O of accelerators. more
| Name | Since | Members |
|---|---|---|
| Memory Management for CXL-Based Memory Architectures | 2025 | Habicht, Daniel |
| CXL-Based Hybrid SSDs | 2023 | Daniel Habicht, Yussuf Khalil |
| Fast and Power-Efficient System Suspend | 2023 | Yussuf Khalil |
| GPU4FS: The Fully GPU-Accelerated File System | 2022 | Peter Maucher |
For information on former projects please contact Prof. Dr.-Ing. Frank Bellosa.