Elastic In-Memory Transaction Processing for Multi-Tenant Database Systems
Event details
| Date | 05.06.2018 |
| Hour | 13:00 › 15:00 |
| Speaker | Seyedeh Sharareh Mirzargar |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Edouard Bugnion
Thesis advisor: Prof. Anastasia Ailamaki
Co-examiner: Prof. James Larus
Abstract
Online Transaction Processing (OLTP) is one of the most important database applications. In recent years, the emergence of new internet-scale companies (e.g., Facebook), has increased the need for ever higher transactional throughput. For example, OLTP deployments at Facebook process over 170 million transactions each second. This new OLTP market has time-varying workloads due to daily, weekly or seasonal fluctuations in demand. Despite this highly variable and diverse load, OLTP database management systems need to maintain their high throughput. Thus, in order to support these applications, OLTP DBMS has recently started to migrate from on-premise setting to the cloud infrastructure to benefit from the available elasticity.
In this proposal, we will first demonstrate how in-memory transaction processing has successfully provided high throughput for OLTP engines by making fundamental changes to the system architecture. Subsequently, we discuss a fine-grained repartitioning framework which provides elasticity for on-premise shared-nothing OLTP DBMS. Then, we will discuss elastic transaction processing in multi-tenant cloud infrastructure. We present SQLVM which provides an abstraction of a virtual machine inside the DBMS to isolate the performance of multiple tenants from each other with low overhead. Finally, we conclude with our research proposal regarding elastic in-memory transaction processing for multi-tenant scale-up OLTP deployment.
Background papers
The End of an Architecture Era (It's Time for a Complete Rewrite) In the proceedings of the 33rd International Conference on Very Large Databases (VLDB'07), by Stonebraker M., et al.
E-Store: Fine-Grained Elastic Partitioning for Distributed Transaction Processing Systems. In the proceedings of the 41st International Conference on Very Large Databases (VLDB'15), by Taft R., et al.
SQLVM: Performance Isolation in Multi-Tenant Relational Database-as-a-Service. In 6th Biennial Conference on Innovative Data Systems Research(CIDR'13), by Narasayya V., et al.
Exam president: Prof. Edouard Bugnion
Thesis advisor: Prof. Anastasia Ailamaki
Co-examiner: Prof. James Larus
Abstract
Online Transaction Processing (OLTP) is one of the most important database applications. In recent years, the emergence of new internet-scale companies (e.g., Facebook), has increased the need for ever higher transactional throughput. For example, OLTP deployments at Facebook process over 170 million transactions each second. This new OLTP market has time-varying workloads due to daily, weekly or seasonal fluctuations in demand. Despite this highly variable and diverse load, OLTP database management systems need to maintain their high throughput. Thus, in order to support these applications, OLTP DBMS has recently started to migrate from on-premise setting to the cloud infrastructure to benefit from the available elasticity.
In this proposal, we will first demonstrate how in-memory transaction processing has successfully provided high throughput for OLTP engines by making fundamental changes to the system architecture. Subsequently, we discuss a fine-grained repartitioning framework which provides elasticity for on-premise shared-nothing OLTP DBMS. Then, we will discuss elastic transaction processing in multi-tenant cloud infrastructure. We present SQLVM which provides an abstraction of a virtual machine inside the DBMS to isolate the performance of multiple tenants from each other with low overhead. Finally, we conclude with our research proposal regarding elastic in-memory transaction processing for multi-tenant scale-up OLTP deployment.
Background papers
The End of an Architecture Era (It's Time for a Complete Rewrite) In the proceedings of the 33rd International Conference on Very Large Databases (VLDB'07), by Stonebraker M., et al.
E-Store: Fine-Grained Elastic Partitioning for Distributed Transaction Processing Systems. In the proceedings of the 41st International Conference on Very Large Databases (VLDB'15), by Taft R., et al.
SQLVM: Performance Isolation in Multi-Tenant Relational Database-as-a-Service. In 6th Biennial Conference on Innovative Data Systems Research(CIDR'13), by Narasayya V., et al.
Practical information
- General public
- Free
Contact
- EDIC - [email protected]