Data-Intensive Distributed Systems LaboratoryCS Seminar
Date: January 21st, 2026
Time: 12:50pm
Room: SB212
Lan Nguyen
PhD student, Computer Science Department, Illinois Institute of Technology.
Talk Title
XStore: A High-performance Lightweight Time-series Storage System
Talk Abstract
Time-series data are growing at an unprecedented rate across industries, driven by pervasive sensing, high-frequency financial markets, and real-time analytics systems. The scale of data generated by financial instruments at fine temporal granularity poses significant challenges to state-of-the-art time-series storage systems, particularly when datasets span billions to trillions of samples. This paper presents XStore, a time-series storage system implemented in C++ that employs a novel design combining sparse files with timestamp mapping to achieve O(1) time complexity for fundamental operations, including create, read, update, and delete. We evaluate XStore on a 10-year dataset with one-second granularity, demonstrating performance improvements under highly concurrent workloads of up to 1013x over MongoDB and 2894x over InfluxDB. XStore's timestamp mapping enables constant-time insertion and lookup while eliminating the need for conventional indexing structures such as memory-intensive tables and complex tree-based designs. As a result, the system exhibits substantially higher resource efficiency, maintaining a memory footprint of 21MB compared to 26GB for competing systems, even under a 64-client load. To assess real-world applicability, we further evaluate XStore using 10 years of financial price data in a backtesting application, benchmarking against MongoDB, InfluxDB, CSV (Pandas), and Parquet. Across 1-512 concurrently running backtesting sessions, XStore achieves average speedups of 3x-8x.
Speaker Bio
Lan Nguyen is a 5th year PhD student in the Computer Science Department at Illinois Institute of Technology working on time-series storage and data management systems.