Design Principles of Hybrid Transactional and Analytical Processing (HTAP) Systems
Event details
Date | 12.06.2019 |
Hour | 14:00 › 16:00 |
Speaker | Aunn Raza |
Location | |
Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Rachid Guerraoui
Thesis advisor: Prof. Anastasia Ailamaki
Co-examiner: Prof. Edouard Bugnion
Abstract
Modern applications generate and update large data volumes. To provide meaningful business intelligence, applications require timely analytics on fresh operational data. Nevertheless, recent work on database management systems has advocated a transition from monolithic engines to separate analytic- and transaction-specialized ones, to lighten up the critical path using workload knowledge. As a result, supporting hybrid transactional and analytical processing (HTAP) requires an out-of-band periodic extract-transform-load (ETL) process to synchronize the two engines. ETL causes a trade-off between analytical data freshness and performance, which becomes more evident as data freshness becomes increasingly vital.
In this proposal, we discuss the design principles of HTAP systems and techniques to reduce the ETL cost through three existing works. The first work aims to provide analytical data freshness by replacing the costly ETL process with lazy snapshotting. The second work aims to provide performance isolation between analytical and transactional workloads through hardware isolation. The third work exploits heterogeneous hardware for scaling hybrid workloads on modern hardware. Finally, we conclude with our vision for adaptive scheduling of hybrid workloads to dynamically balance between isolation and analytical data freshness.
Background papers
HyPer: A hybrid OLTP&OLAP main memory database system based on virtual memory snapshots, by A. Kemper, and T. Neumann. In ICDE, pages195-206, 2011.
BatchDB: Efficient isolated execution of hybrid OLTP+ OLAP workloads for interactive applications,by D. Makreshanski, J. Giceva, C. Barthels, and G. Alonso. In SIGMOD, pages 37-50, 2017.
The case for heterogeneous HTAP, by R. Appuswamy, M. Karpathiotakis, D. Porobic, and A. Ailamaki. In CIDR, 2017.
Exam president: Prof. Rachid Guerraoui
Thesis advisor: Prof. Anastasia Ailamaki
Co-examiner: Prof. Edouard Bugnion
Abstract
Modern applications generate and update large data volumes. To provide meaningful business intelligence, applications require timely analytics on fresh operational data. Nevertheless, recent work on database management systems has advocated a transition from monolithic engines to separate analytic- and transaction-specialized ones, to lighten up the critical path using workload knowledge. As a result, supporting hybrid transactional and analytical processing (HTAP) requires an out-of-band periodic extract-transform-load (ETL) process to synchronize the two engines. ETL causes a trade-off between analytical data freshness and performance, which becomes more evident as data freshness becomes increasingly vital.
In this proposal, we discuss the design principles of HTAP systems and techniques to reduce the ETL cost through three existing works. The first work aims to provide analytical data freshness by replacing the costly ETL process with lazy snapshotting. The second work aims to provide performance isolation between analytical and transactional workloads through hardware isolation. The third work exploits heterogeneous hardware for scaling hybrid workloads on modern hardware. Finally, we conclude with our vision for adaptive scheduling of hybrid workloads to dynamically balance between isolation and analytical data freshness.
Background papers
HyPer: A hybrid OLTP&OLAP main memory database system based on virtual memory snapshots, by A. Kemper, and T. Neumann. In ICDE, pages195-206, 2011.
BatchDB: Efficient isolated execution of hybrid OLTP+ OLAP workloads for interactive applications,by D. Makreshanski, J. Giceva, C. Barthels, and G. Alonso. In SIGMOD, pages 37-50, 2017.
The case for heterogeneous HTAP, by R. Appuswamy, M. Karpathiotakis, D. Porobic, and A. Ailamaki. In CIDR, 2017.
Practical information
- General public
- Free