Parallel Data Laboratory Talks

— 2:00pm

Virtual Presentations - ET - Remote Access - Zoom

12:00 pm  —  Greg Ganger, Director, Parallel Data Laboratory and Professor, Departments of Computer Science and Electrical ad  Computer Engineering, CMU
                             Welcome and Introductions

12:10 pm  —  Andy Pavlo, Associate Professor, Computer Science Department, CMU
                             OtterTune: An Automatic Database Configuration Tuning Service

Database management systems (DBMS) expose dozens of configurable knobs that control their runtime behavior. Setting these knobs correctly for an application's workload can improve the performance and efficiency of the DBMS. But such tuning requires considerable efforts from experienced administrators, which is not scalable for large DBMS fleets. This problem has led to research on using machine learning (ML) to devise strategies to optimize DBMS knobs for any application automatically. The OtterTune database tuning service from Carnegie Mellon uses ML to generate and install optimized DBMS configurations. OtterTune observes the DBMS's workload through its metrics and then trains recommendation models that select better knob values. It then reuses these models to tune other DBMSs more quickly.

Andy Pavlo is an Associate Professor of Databaseology in the Computer Science Department at Carnegie Mellon University. He is also the co-founder of OtterTune.

1:00:pm  —  George Amvrosiadis, Assistant Research Professor, Department of Electrical and Computer Engineering, CMU
                           Getting Zoned Storage out of the friend zone

Zoned Storage is coming to change the way we interact with storage devices. While the simple block interface served us well for decades, modern storage devices have struggled to fit its semantics. We have spent more than a decade building complex translation layers, garbage collection algorithms, and tucking away bits to help mask how different the inner workings of storage devices are, from what is presented to applications. And yet, the ghost of tail latency still haunts us. Zoned Storage proudly exposes these inner workings, ridding applications from the complexity of “compatibility” but letting them figure out how to make do with constraints of the media. We’re here to help identify use cases that make sense for zoned devices, and make the transition (from the block interface) smoother. I’ll discuss how we made key-value stores work with zoned storage devices, contributing patches to RocksDB and the Linux kernel, and our plans to make zones fit into RAID configurations and data caching systems.

George Amvrosiadisis an Assistant Research Professor of Electrical and Computer Engineering at Carnegie Mellon University. His current research focuses on distributed and cloud systems infrastructure, new storage technologies, high-performance computing, and systems for machine learning. His group’s research has received an R&D100 Award, and has been featured on WIRED, The Morning Paper, Hacker News.

Event Website:

For More Information: