Scalable transactional processing through operations-aware protocols
Event details
| Date | 25.11.2024 |
| Hour | 10:00 › 12:00 |
| Speaker | Ioanna Tsakalidou |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Sanidhya Kashyap
Thesis advisor: Prof. Anastasia Ailamaki
Co-examiner: Prof. Anne-Marie Kermarrec
Abstract
Modern Online Transaction Processing (OLTP)
systems require concurrency control protocols to maintain
consistency and isolation in the database while interleaving
concurrent operations to improve throughput.
In this research proposal, we outline three different concurrency
protocols for transaction execution. The first protocol,
called the healing protocol, optimizes Optimistic Concurrency
Control (OCC) by addressing the costs associated
with aborting and restarting transactions, exploiting the sets
of transactions’ operations to resolve inconsistencies in the
state of invalidated transactions. The second protocol, the
data partitioning protocol, partitions the database and employs
fine-grained locking to reduce contention, while it
executes transactions using different schemes based on the
number of partitions they access. The third protocol, the
workload clustering protocol, minimizes unnecessary concurrency
control (CC) overheads, such as locking or actions
for validating transactions, and is suitable for dynamically
varying workloads, even with multi-partition transactions,
by clustering non-conflicting transactions and allowing them
to execute in parallel.
Finally, inspired by these approaches, we briefly present
our vision for adapting the concurrency control protocol and
the granularity of conflict detection in oltp workloads.
Background papers
Exam president: Prof. Sanidhya Kashyap
Thesis advisor: Prof. Anastasia Ailamaki
Co-examiner: Prof. Anne-Marie Kermarrec
Abstract
Modern Online Transaction Processing (OLTP)
systems require concurrency control protocols to maintain
consistency and isolation in the database while interleaving
concurrent operations to improve throughput.
In this research proposal, we outline three different concurrency
protocols for transaction execution. The first protocol,
called the healing protocol, optimizes Optimistic Concurrency
Control (OCC) by addressing the costs associated
with aborting and restarting transactions, exploiting the sets
of transactions’ operations to resolve inconsistencies in the
state of invalidated transactions. The second protocol, the
data partitioning protocol, partitions the database and employs
fine-grained locking to reduce contention, while it
executes transactions using different schemes based on the
number of partitions they access. The third protocol, the
workload clustering protocol, minimizes unnecessary concurrency
control (CC) overheads, such as locking or actions
for validating transactions, and is suitable for dynamically
varying workloads, even with multi-partition transactions,
by clustering non-conflicting transactions and allowing them
to execute in parallel.
Finally, inspired by these approaches, we briefly present
our vision for adapting the concurrency control protocol and
the granularity of conflict detection in oltp workloads.
Background papers
- Wu, Y. and Chan, C.-Y. and Tan, K.-L., Transaction Healing: Scaling Optimistic Concurrency Control on Multicores, In SIGMOD, pages 1689–1704, 2016
- Zhou, X. and Yu, X. and Graefe, G and Stonebraker, M., Lotus: Scalable Multi-Partition Transactions on Single-Threaded Partitioned Databases, In VLDB, pages 2939-2952, 2022
- Prasaad, G. and Cheung, A. and Suciu, D., Handling Highly Contended OLTP Workloads Using Fast Dynamic Partitioning, In SIGMOD, pages 527–542, 2020
Practical information
- General public
- Free