Data Storage Lab

Our research are motivated by problems of mission-critical systems that jeopardize data, e.g.:

Crash, Corruption, & Bug across system layers ==> [SSD: TOCS'16/FAST'13 | OS: TOS'23 | FS: FAST'23, TOS'18/FAST'18 | DB: OSDI'14] Data Lost & Service Disruption @HPC/Data centers at scale ==> [HotStorage'24, TOS'22/ICS'18, HotStorage'21, ATC'19] Data Provenance, Observability, & Scalability challenges ==> [TPDS'24/HPDC'22, TBench'21, ASPLOS'23, ARA Platform]

We build systems/tools to attack such problems and open source our research prototypes/datasets.

Accepting new Ph.D. students for Spring'25/Fall'25! Please email if interested!

• Current Members
  
  
  • Faculty: Mai Zheng
  • Students:

    Tabassum Mahmud	Wei Xu	Chao Shi	Varun S. Girimaji	Joshua John

    Vidhya Mannathu Parambil	Roop Kiran	Jashwanth Kumar Komanabelli	Zeren Yang

  
  
• Selected Projects
  
  
  • Large-Scale High-Performance Storage Systems
    

    Large-scale distributed storage systems (DSS) are deployed widely in the world to empower data-intensive computing. The shift towards data-driven discoveries demands high-performance DSS that can manage massive datasets efficiently and reliably. Nevertheless, state-of-the-art DSS may experience correctness and/or performance issues due to the scale and complexity. We are building frameworks to improve data storage for robust data-driven discoveries at scale. Details: [HotStorage'24 | ASPLOS'23 | IPDPS'23a | IPDPS'23b | TOS'22 | HotStorage'21 (video) | PDSW'20 (slides) | MSST'19 | ICS'18 (slides) | PDSW'16 (slides)] Open Source Artifact: [PFault]

  
  
  • Local File Systems & Utilities

    Local file systems (e.g., Ext4, ZFS) are the cornerstone of many systems and applications today. Unfortunately, despite decades of evolution and various protections, local file systems may still run into issues in practice (e.g., data corruptions, metadata inconsistencies). What's worse, the same issues that occur to file systems may also affect their maintenance utilities (e.g., e2fsprogs) and/or other applications using them, leading to cascading problems. We are investigating the problems and developing mitigation solutions. Details: [FAST'23 | HotStorage'22 (slides) | FAST'22-WiP (video) | TOS'18 | FAST'18 (slides) | HotStorage'17 (slides) | FAST'17-WiP] Open Source Artifact: [ConfD] [rFSCK]

  
  
  • Implications of Non-Volatile Memories
    

    Non-volatile memory (NVM) technologies, including flash-based solid-state drives (SSD) and byte-addressable persistent memories (PM), are disrupting the storage market. Modern storage systems increasingly use SSD or PM for high performance and low energy cost. Nevertheless, building correct NVM storage systems is non-trivial due to unique device characteristics. We take a holistic view to analyze both NVM devices and host software to understand and improve the overall system robustness and performance. Details: [TOS'23 | NVMW'23 | TOC'22 | SYSTOR'21 (video) | FAST'19-WiP | ATC'19 (slides) | PDSW'18 (slides) | TOCS'16 | FAST'13 (slides)] Open Source Artifact: [BugBenchk]

  
  
  • Full-Stack/Cross-Layer Issues in Storage
    

    The storage stack (i.e., devices, drivers, file systems, databases, distributed data management systems) is complicated. Besides potential issues in individual layers, many desired properties may be violated due to external failure events (e.g., power outages), configuration dependencies, etc., hurting the end-to-end data integrity. We take a holistic view to investigate such daunting challenges, and many of our works involve more than one single layer. Details: [TOS'23 | FAST'23 | Dagstuhl'22 | TOS'22 | ATC'19 | TOS'18 | TOCS'17 | OSDI'14 (slides) ] Open Source Artifacts: [ConfD] [rFSCK] [PFault]

  
  
  • Data Provenance & Observability
    

    Understanding the origin and quality of data (e.g., the root cause of an anomaly, the reproducibility of the best result) becomes more and more challenging due to the ever-increasing data volume and system complexity. Data provenance, or data lineage, describes the life cycle of data, which is essential to address the challenge. We are building frameworks to capture rich metadata and generate provenance to ensure the observability, reproducibility, explainability, auditability, trustworthiness, etc of systems and data. Details: [TPDS'24 | HPDC'22 | NAS'22 | FAST'22-WiP | DOE-ASRC-Data'22 | TBench'21 (video) | HotStorage'20 (slides) | FAST'20-WiP] Open Source Artifacts: [PROV-IO] [BugBenchk]

  
  
  • Data Infrastructure for Wireless & Rural America
    

    ARA is an at-scale data infrastructure for advanced wireless research being deployed across the ISU campus, City of Ames, and surrounding research and producer farms as well as rural communities in central Iowa, spanning a rural area with diameter over 60km. It serves as a wireless living lab for smart and connected rural communities and rich experimental data, enabling the research and development of rural-focused wireless technologies that provide affordable, high-capacity connectivity to rural communities and industries such as agriculture. Details: [MobiCom'23-Demo | WiNTECH'22 | WiNTECH'21] Infrastructure: [ARA]

  
  


• Selected Outreach Activities
  
  
  
  
  
  
  • 
    
    
• Full Publication List
  
• Alumni
  
• Acknowledgements