Computer Systems and Engineering Seminar Series
What Non-Volatile Memory Means For Database Systems?
||Friday, November 3, 2017
||12:00pm - 1:00pm
||D344 LSRC, Duke
For the first time in 25 years, a new non-volatile memory (NVM) category is being created that is two orders of magnitude faster than current durable storage devices. The advent of NVM fundamentally changes the dichotomy between memory and durable storage in database systems. These new NVM devices are almost as fast as DRAM, but all writes to it are potentially persistent even after power loss. Existing database systems are unable to take full advantage of this technology because their internal architectures are predicated on the assumption that memory is volatile. With NVM, many of the components of legacy database systems are unnecessary and will degrade the performance of applications in several modern data-intensive domains.
In this talk, I will discuss the implications of NVM for database systems. Specifically, I will introduce a new logging and recovery algorithm, called write-behind logging that increases the availability of the system by two orders of magnitude compared to the state-of-the-art write-ahead logging algorithm. Write-behind logging leverages the unique characteristics of NVM to improve the operational cost and performance of the database system. I will also present the design of a new indexing data-structure that runs seamlessly on both DRAM and NVM without requiring special-purpose recovery code, which greatly reduces the code development and maintenance costs. In drawing broader lessons from this work, I believe that these algorithms and data-structures significantly simplify the development of high-performance software systems for emergent NVM technologies.
Joy Arulraj is a Ph.D. candidate in the Parallel Data Lab at Carnegie Mellon University working in database systems. As part of his dissertation work, he has studied and built a new non-volatile memory database system, called Peloton, for large-scale transaction processing and real-time data analytics under the advisement of Andrew Pavlo, Garth Gibson, and Todd Mowry. His research is supported by the Carnegie Mellon Presidential Fellowship (2018) and the Samsung Fellowship (2017).
Hosted by: Dr. Ashwin Machanavajjhala