Context Engineering Paradigm Shift

From storage optimization to context orchestration. IOWarp treats data + metadata + computational state as unified context for autonomous agents.

The Fundamental Paradigm Shift

For decades, scientific computing optimized for data throughput and storage purity. But in the era of autonomous scientific AI, this optimization target is fundamentally misaligned. Agents don't need raw bytes—they need context. The paradigm must shift from optimizing storage systems for human scientists to orchestrating context for autonomous agents.

The Data Paradox

175 zettabytes by 2025, yet AI is "data-starved." 80% of scientific data is unstructured or siloed. Missing semantic relationships. Format fragmentation. IOWarp bridges this gap by treating data, metadata, and computational state as unified context—enabling agents to reason, not just retrieve.

Research Evolution

2018

Hermes

Multi-tiered storage optimization, automated data movement

2019

LABIOS

Label-based I/O, semantic-aware placement

2020

ChronoLog

Time-series aware storage

2024

MegaMmap

Memory-storage blending (SC'24 paper)

2024

IOWarp

Full context orchestration for agentic AI

2025

Context Engineering

New discipline for autonomous discovery

Key Research Contributions

MegaMmap (SC'24)

Memory-storage blending architecture enabling seamless data movement between DRAM and storage tiers.

I/O Performance 20% faster
DRAM Reduction 2.6x

Hermes

Multi-tiered buffering system with ML-driven prefetching for HPC workloads.

Throughput 3x improvement
Deployed 5+ national labs

LABIOS

Label-based asynchronous I/O system enabling semantic-aware data placement and retrieval.

Format Support 50+ formats
Metadata Extraction Automatic

GenomIO

Genomics I/O optimization framework for variant analysis and genome-wide association studies.

Speedup 10x
Scale Petabyte datasets

ChronoLog

Distributed time-series storage system optimized for temporal queries and sensor data fusion.

Query Performance 100x faster
Compression 5x reduction
Real-time Sub-second latency
Deployment EarthScope

Recognition

2
Best Paper Awards
1
US Patent
$15M
Cumulative Funding

Demo Highlights

IOWarp Publications (15)

All Storage Optimizations AI Agents HPC Systems Context Engineering

Jarvis: Towards a Shared, User-Friendly, and Reproducible I/O Infrastructure

2024
J. Cernuda, L. Logan, N. Lewis, S. Byna, X.-H. Sun, A. Kougkas
The International Parallel Data Systems Workshop (PDSW'24) storage-optimizations

Uncover the Overhead and Resource Usage for Handling KV Cache Overflow in LLM Inference

2024
J. Ye, B. Nicolae, A. Kougkas, X.-H. Sun
The International Conference for High Performance Computing, Networking, Storage, and Analysis (SC'24) storage-optimizations

MegaMmap: Blurring the Boundary Between Memory and Storage for Data-Intensive Workloads

2024
Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis SC storage-optimizations
View Publication →

DFTracer: An Analysis-Friendly Data Flow Tracer for AI-Driven Workflows

2024
Proceedings of the International Conference for High Performance Computing, Networking, Storage, and Analysis agent-intelligence
View Publication →

To Derive or Not to Derive: I/O Libraries Take Charge of Derived Quantities Computation

2024
2024 IEEE 36th International Symposium on Computer Architecture and High Performance Computing (SBAC-PAD) core-infrastructure
View Publication →

DaYu: Optimizing Distributed Scientific Workflows by Decoding Dataflow Semantics and Dynamics

2024
Proceedings of the IEEE International Conference on Cluster Computing core-infrastructure
View Publication →

An Evaluation of DAOS for Simulation and Deep Learning HPCWorkloads

2024
Proceedings of the ACM SIGOPS Operating Systems Review core-infrastructure
View Publication →

Viper: A High-Performance I/O Framework for Transparently Updating, Storing, and Transferring Deep Neural Network Models

2024
Proceedings of the 53rd International Conference on Parallel Processing storage-optimizations
View Publication →

HStream: A hierarchical data streaming engine for high-throughput scientific applications

2024
Proceedings of the 53rd International Conference on Parallel Processing core-infrastructure
View Publication →

TunIO: An AI-powered Framework for Optimizing HPC I/O

2024
Proceedings of the International Parallel and Distributed Processing Symposium agent-intelligence
View Publication →

Data Flow Lifecycles for Optimizing Workflow Coordination

2023
Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis core-infrastructure
View Publication →

IOMax: Maximizing Out-of-Core I/O Analysis Performance on HPC Systems

2023
Proceedings of the SC'23 Workshops of The International Conference on High Performance Computing, Network, Storage, and Analysis storage-optimizations
View Publication →

An Evaluation of DAOS for Simulation and Deep Learning HPC Workloads

2023
Proceedings of the 3rd Workshop on Challenges and Opportunities of Efficient and Performant Storage Systems core-infrastructure
View Publication →

LabStor: A modular and extensible platform for developing high-performance, customized I/O stacks in userspace

2022
Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis core-infrastructure
View Publication →

Stimulus: Accelerate Data Management for Scientific AI applications in HPC

2022
Proceedings of the 22nd IEEE International Symposium on Cluster, Cloud and Internet Computing storage-optimizations
View Publication →

Showing IOWarp project publications. For the full GRC research portfolio (25+ papers), visit grc.iit.edu/publications.

Active Research Areas

Context Engineering

Novel approaches to representing and orchestrating context for autonomous agents. Hierarchical memory systems, semantic indexing, and context-aware prefetching.

Agent Orchestration

Multi-agent coordination frameworks, shared context protocols, and workflow optimization for scientific discovery pipelines.

Storage Systems

Multi-tier storage optimization, CXL/RDMA integration, and intelligent data placement strategies for petabyte-scale workloads.

Domain Applications

Specialized optimizations for climate science, materials discovery, genomics, seismology, and astrophysics applications.

Collaboration Opportunities

Open to academic, industry, and government partnerships. We welcome collaborations on context engineering, agentic science infrastructure, and domain-specific applications.

Interested in contributing? We offer research internships, visiting scholar positions, and collaborative grant opportunities.