Sky Seminar: Ram Alagappan (UIUC) – New Log Abstractions for Datacenter Applications


Speaker: Ram Alagappan
Location: Soda 510
Date: September 27th, 2024
Time: 12pm-1pm PDT

Title: New Log Abstractions for Datacenter Applications

Abstract:
The log is arguably the simplest yet most pervasive storage abstraction at the heart of many applications. For example, traditional storage-centric applications like databases and key-value stores are built around logs. Similarly, modern applications like stream processing and event sourcing are built atop shared logs (like Kafka). Despite its widespread use, the log can present several performance challenges to applications.    

Our group has been building new log abstractions to address this challenge for traditional and modern applications. The first abstraction, Near-Compute Logs (EuroSys 24), enables traditional storage-centric applications like databases running atop disaggregated cloud storage to realize strong fault-tolerance guarantees without forgoing performance. The second one, LazyLog (SOSP 24), is a new shared-log abstraction better suited for applications that demand low-latency ingestion. In this talk, I will mostly focus on LazyLog, describing its motivation, design, and benefits. 

Bio:
Ram Alagappan is an Assistant Professor at UIUC, where he co-leads the Distributed And Storage Systems Lab (DASSL). His research focuses on storage systems, disaggregated memory, and distributed systems. His work has appeared at OSDI, SOSP, FAST, and EuroSys. He has won several awards, including an NSF CAREER award, teaching recognitions at UIUC, and best-paper awards at FAST ’20, FAST ’18, and FAST ’17. His open-source tools have had a practical impact, exposing more than 80 severe crash vulnerabilities across 20 widely used systems. Ideas from his work have been adopted by a financial database startup.