Abstract
Nowadays, in-memory computing has plenty of applications like artificial intelligence, databases, machine learning, etc. These applications usually involve with the frequent access to memory. On the other hand, memory components typically become error-prone over time due to the increase of density and capacity. It is urgently important to develop solutions for high-availability memory access. Yet, existing solutions are either lack of flexibility, or consistently more expensive than native memory. To the end, this paper presents a solution called SC2M. It is a software-controlled, high-availability memory mirroring solution. Our solution can flexibly set the granularity of the memory areas for various levels. Furthermore, it can perform duplication of the user-defined data structures in a high-availability version. The systematic instruction-level granularity for memory duplication reduces the overheads for backup, and lowers the probability of data loss. Experiment results demonstrate the feasibility and superiorities of our solution.
Original language | English |
---|---|
Pages | 122-137 |
Number of pages | 16 |
DOIs | |
Publication status | Published - 2020 |
Event | 19th International Conference on Algorithms and Architectures for Parallel Processing, ICA3PP 2019 - Melbourne, Australia Duration: 9 Dec 2019 → 11 Dec 2019 |
Conference
Conference | 19th International Conference on Algorithms and Architectures for Parallel Processing, ICA3PP 2019 |
---|---|
Country/Territory | Australia |
City | Melbourne |
Period | 9/12/19 → 11/12/19 |
Keywords
- Hardware virtualization
- High availability
- System architecutre
ASJC Scopus subject areas
- Theoretical Computer Science
- General Computer Science