Research Projects

A subset of the current projects are listed below, with links to a representative paper for that project. More details can be found in the projects page of the TechInt group website and in my vita.

• Deep-Storage Hierarchy Projects
  • Spider Storage Design & Deployment
  • I/O Signature Extraction
  • Parallel File System Metadata Analysis
  • TagIt: An integrated search and discovery service for file systems
  • Data Jockey: Tiered Storage Management
  • Understanding User Behavior and Data Sharing Trends in a parallel file system
  • HPSS Archival Storage
  • SSD-based Burst Buffer Storage for Summit: SymphonyFS (combining node-local SSD on Summit and GPFS into a single FS), Spectral (seamless use of node-local SSD)
  • Active Flash
  • AnalyzeThis: Analysis-aware Storage System
• Deep-Memory Hierarchy Projects
  • NVMalloc: SSD-based Memory Extension
  • AnalyzeThat: A Programmable Shared-Memory System for Processing-In-Memory Devices
  • Understanding Memory Object Access Patterns
• Data, Logs, Metadata Management and Mining
  • Constellation Science Graph Network for Scalable Data Discovery
  • DOIs for HPC Data
  • Graph Analytics to Understand Compute Core Usage and Publication Trends
  • GUIDE—Data Analytics on large-scale Logs to Optimize HPC Data center Operations
• Computer System Architecture and Resilience
  • Functional Partitioning Runtime for Heterogeneous nodes
  • Supercomputer Reliability Analysis
  • GPU Age-aware Scheduling
  • Scheduling to reduce fragmentation, and improve communication BW and reliability
  • System Provisioning and Projections
• OLCF4 Summit Design, Deployment and Evaluation
• OLCF5 Exascale Procurement

Selected Previous Work

• Staging Architecture using non-volatile devices to Accelerate I/O Pipelines on HPC systems
• stdchk: A Checkpoint Storage System for HPC Applications
• FreeLoader: A client-side cache for scientific data using distributed storage scavenging on desktop machines
• Transient Data Recovery in HPC Centers
• /Scratch As a Cache
• End-User Data Delivery
• Bringing Neutron Science data and simulations to the TeraGrid
• Bulk Data Transfer Forecasts and the Implications to Grid Scheduling - Ph.D. Dissertation Work
  • This involved research and development on Data Grid tools and techniques in conjunction with Argonne's Distributed Systems Laboratory. My doctoral work was based on Globus Data Grid - built techniques to locate data in highly replicated environments; developed performance prediction strategies for bulk Data Grid transfers aiding data transfer schedules; developed co-allocation middleware enabling parallel and fast access to replicated data, etc.