Efficient Analytical Query Processing on Heterogeneous Memory Platforms

Event details
Date | 02.07.2025 |
Hour | 10:00 › 12:00 |
Speaker | Georgiy Lebedev |
Location | |
Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Rachid Guerraoui
Thesis advisor: Prof. Anastasia Ailamaki
Thesis co-advisor: Prof. Sanidhya Kashyap
Co-examiner: Prof. Thomas Bourgeat
Abstract
Modern analytical workloads heavily rely on in-memory processing. However, memory fails to scale to the demands of data-intensive applications. This has led to the advent of new memory technologies, enabling tradeoffs between capacity, cost, latency and bandwidth, and making the memory landscape increasingly heterogeneous. This new norm of memory heterogeneity will have to be deal with by software.
In this proposal, we discuss the adaptive approaches required for solving the challenges in efficient heterogeneous memory management design for analytical processing. The first work provides an adaptive approach to page hotness classification to handle non-uniformity of memory access distributions. The second work presents an adaptive approach to page placement under varying memory access latency. The third work describes a cache management approach for analytical processing systems to adapt to the variation of query characteristics.
Finally, based on the insights from previous approaches, we briefly present our vision for an analytical processing system for heterogeneous memory that improves memory efficiency by leveraging the memory semantics of query execution operators.
Selected papers
1. Demystifying CXL Memory with Genuine CXL-Ready Systems and Devices
2. HPCache: memory-efficient OLAP through proportional caching revisited
3. How to Be Fast and Not Furious: Looking Under the Hood of CPU Cache Prefetching
Exam president: Prof. Rachid Guerraoui
Thesis advisor: Prof. Anastasia Ailamaki
Thesis co-advisor: Prof. Sanidhya Kashyap
Co-examiner: Prof. Thomas Bourgeat
Abstract
Modern analytical workloads heavily rely on in-memory processing. However, memory fails to scale to the demands of data-intensive applications. This has led to the advent of new memory technologies, enabling tradeoffs between capacity, cost, latency and bandwidth, and making the memory landscape increasingly heterogeneous. This new norm of memory heterogeneity will have to be deal with by software.
In this proposal, we discuss the adaptive approaches required for solving the challenges in efficient heterogeneous memory management design for analytical processing. The first work provides an adaptive approach to page hotness classification to handle non-uniformity of memory access distributions. The second work presents an adaptive approach to page placement under varying memory access latency. The third work describes a cache management approach for analytical processing systems to adapt to the variation of query characteristics.
Finally, based on the insights from previous approaches, we briefly present our vision for an analytical processing system for heterogeneous memory that improves memory efficiency by leveraging the memory semantics of query execution operators.
Selected papers
1. Demystifying CXL Memory with Genuine CXL-Ready Systems and Devices
2. HPCache: memory-efficient OLAP through proportional caching revisited
3. How to Be Fast and Not Furious: Looking Under the Hood of CPU Cache Prefetching
Practical information
- General public
- Free
Contact
- edic@epfl.ch