Breaking the Memory Barrier: How EloqKV Replaces DRAM-Based Redis with SSD

Note: please fill out this simple form in advance of attending the meeting to allow for smoother flow at the front door, and help us with organizing this and upcoming events. Thank you!

Abstract:
Driven by AI workloads, rapidly rising DRAM costs are pushing infrastructure toward NVMe and SSD-based designs. However, preserving sub-millisecond tail latency on persistent storage remains a major engineering challenge.
In this session, we explore the architectural patterns required to maintain P99.99 performance while moving beyond memory-only systems. We’ll dive into EloqKV’s execution model, focusing on how coroutines and io_uring are used to manage asynchronous I/O without blocking.
You’ll learn how cache-miss requests are handled efficiently, and the key trade-offs involved in scheduling requests across modern hardware hierarchies.

Speakers:
Jeff Chen is the Co-founder and Chief Architect of EloqData. Prior to EloqData, he was Principal Researcher at Microsoft Research. His research focused on data management systems and data-intensive applications. He was the architect of Azure Cosmos DB Graph, the inventor of SQL Server Selective XML Index and shipped several core innovations to the Bing big data platform. He has authored dozens of publications in top-tier database conferences (SIGMOD, VLDB and ICDE). He holds a PhD in Computer Science from UC San Diego and lives at the intersection of deep research and real-world engineering.

Timing:
7:00--7:30 Social Session
7:30--7:45 Announcements & Introductions
7:45-ish Presentation