The group at OSC has focused on integrative services that include interface design, volume visualization, and parallel volume rendering. Our current visualization cluster uses Myrinet 2000 as network interconnect. Myrinet 2000 peeks out at around 250MBs per second which limits us to around 40 nodes. We are currently exploring PCI-Express which the next generation general I/O connectivity architecture. PCI-Express will replace regular PCI and AGP slots in IA32 and IA64 based computers.
The server chipset (compositing node) will have PCI-Express 8X interfaces. The soon to be released PCI-Express 8X Infiniband card will be capable of 2GBs per second bi-directional bandwidth. Additionally, the workstation chipset (rendering node) will have a PCI-Express 16X slot for graphics. The chipset is capable of 4GBs per second bi-directional bandwidth along with a reduced latency compared with our current AGP4X graphics cards. We are negotiating with industry vendors including Nvidia, Intel, and Mellanox (who is producing an InfiniBand, PCI-Express board) to loan us the beta hardware to run parallel rendering benchmarks. With this new architecture, we intend to scale for use over 100 nodes and interactively render 20 GB's of data.
In addition to the above, the following sections note improvements in OSC's infrastructure that will be exploited by the Center in the near future.
To further evaluate the Center's developments in grid services and grid-based computing, we will exploit OSC's new statewide fiber optic network, the Third Frontier Network (TFN), currently being deployed.
With over 1600 miles of fiber optics (OC-192), TFN will directly link Ohio's colleges and universities to research facilities, businesses, and hospitals and will become the most advanced state-wide computer network in the nation. The goals of TFN are to facilitate new research, collaborations, and course sharing, and to improve Enterprise Resource Planning (ERP) and disaster recovery.
Specific to this proposal, TFN will include private high-speed networks to facilitate instrument sharing, Grid computing, remote collaborations, and data mining. OSC has received a federal appropriation from the DHHS to connect all medical schools and the three Children's Hospital's (Cleveland, Columbus, and Cincinnati) to the TFN. In addition, all Wright Centers of Innovation (WCI's) throughout Ohio will be linked through the TFN to share expensive instruments and educational resources for disease analysis and treatment. The WCI's include specific studies in the biosciences, including Stem Cell Research (Case Western Reserve University, Cleveland Clinic Foundation, University of Cincinnati) and The Biomedical Structural, Functional, and Molecular Imaging Enterprise (The Ohio State University, Case Western Reserve University). TFN will be connected to the National LamdaRail in the Autumn of 2004.
In addition, the Center will utilize Cluster Ohio, an initiative of OSC and the Ohio Board of Regents. OSC has successfully deployed 2nd generation cluster computing systems throughout the state , including 178 Itanium-32 processors through competitive awards made to Wilberforce University, Bowling Green State University, University of Dayton, University of Toledo, Youngstown State University, Ohio University, Kent State University, Wright State University, University of Akron, and Case Western Reserve University. The second round of Cluster Ohio (Spring '03) disseminated 144 Itanium-64 CPU's in designated clusters of 4-16 node, and 1-8 node cluster. Round 3, in the Spring of 2004, disseminated 52 CPU's (32-bit AMD Athlon) to 5 sites statewide (Ohio State University (2), Case Western Reserve University, University of Toledo, and Ohio University). This unique experimental network and cluster environment offers a unique opportunity to facilitate dissemination and evaluation of the Center's developments and to increase participation from the statewide and national community.
To accommodate the increasingly large data intensive activities of this and other projects throughout the state, OSC has installed new Mass Storage Equipment (MSE). This system is comprised of 2 Cisco MDS-9509 FC/iSCSI switches and 4 IBM FAStT-900 FC RAID controllers with 4 FC links per controller, for a total of approximately 50 TB raw capacity. In addition, we have installed 20 IBM FAStT-600 FC RAID controllers, with 2 FC links per controller, for a total of approximately 400 TB raw capacity. We have installed an IBM 3584 LTO tape robot, with 400 Ultrium/LTO tapes, providing a total of approximately 80 TB of capacity. The system runs 26 IBM xSeries servers.