Scalable Transaction Processing with High Contention

Online transaction processing (OLTP) is critical for modern Internet applications across various domains. It is challenging to scale OLTP databases to massive parallelism when the workload is highly contentious. Concurrency control is a major scalability bottleneck in these systems.

This talk covers several projects that identify and improve scalability of databases with high contention. I will present a scalability study of concurrency control on a simulated 1000-core processor and show the bottlenecks that constrain the scaling of classic algorithms [VLDB'14]. Then, I present a protocol called TicToc [SIGMOD'16] that removes the bottleneck of central timestamp allocation on multicore processors and its extension to a distributed concurrency control protocol called Sundial [VLDB'18]. Finally, I will present our recent work Bamboo [SIGMOD'21] that improves conventional two-phase locking to mitigate hotspots.

Please email for a
Zoom link

Who
Xiangyao Yu
Where
Online
Presenter Bio
Xiangyao Yu is an Assistant Professor at the University of Wisconsin-Madison working on database systems, with a focus on (1) transactions and HTAP, (2) new hardware for databases (e.g., GPU, NVM, RDMA, SmartNIC), and (3) cloud-native databases. He finished his postdoc in 2019 with Prof. Michael Stonebraker and Prof. Samuel Madden and received PhD degree in 2017 with Prof. Srinivas Devadas at CSAIL MIT and bachelor degree in 2012 at Tsinghua University.
Speaker affiliation
University of Wisconsin-Madison