Learn more about how Ohio researchers are successfully applying high-performance computing to their research.
OSC engages in research on multiple aspects of high-performance storage and file systems. One NSF-funded project is underway to look at the suitability for a new type of "smart" disks, called object-based storage devices (OSDs), for use in parallel file systems. Topic areas include mapping data to objects, metadata, transport, caching and reliability. Trade-offs arise from the mapping of files to objects, and how to stripe files across multiple objects and disks, in order to obtain good performance.
OSC is a major contributor to the Parallel Virtual File System (PVFS) project, working with colleagues at Argonne National Labs, Clemson University, and other research and industrial partners. Staff at OSC wrote and maintain the RDMA-based transport interface for PVFS, enabling it to use InfiniBand and other high-performance networks. A related project looks at distributed metadata algorithms, using PVFS as a research testbed so that the broader community of PVFS users can take direct advantage of the research efforts.
Another area of research involves using using modern high-speed networking devices to improve Internet-scale communications. In particular, staff and students at OSC have improved the performance and scalability of the popular Apache web server by using iWARP devices. These new ethernet cards communicate directly with existing ethernet infrastructure, but offer the benefits of RDMA communications. This extends the use of recently developed networking techniques in the wide-area.
- PVFS in the WAN
- Applicability of Object-Based Storage Devices in Parallel File Systems
- RDMA Enabled Apache (mod_rdma)
- Software implementation and testing of iWarp protocol
- Low-latency gigabit ethernet message passing
- High Performance and Scalable MPI Implementation on InfiniBand
- Collective Communication and Connection Management Issues in InfiniBand-Based Clusters
- Analysis of message passing environments on large cluster performance
- Supporting MPI collective communication operations with application bypass
- Collective communication and MPICH2 improvements for MVAPICH MPI on InfiniBand
- Job scheduling for petaflop parallel systems
- Database repository using federated databases
- Robust High Performance Parallel Filesystem
- Evaluation of the influence of network topology and mapping on the performance of parallel applications on clusters
Director of Supercomputer Services