Visual Computing Cluster 
 

 

    		•

    Project Overview

    The Stony Brook Visual Computing Cluster was built for two main purposes: (1) as a visualization cluster; and (2) as a computational GPU cluster.  The Labs have a 66-node high-end PC Linux/Windows dual boot cluster with a gigabit Ethernet frontend network and a 10 Gbps InfiniBand backend network, and a portion of computing/display nodes is connected to HP ServerNet high-speed (180MB/s bandwidth) network through HP Sepia-2A card.  Each node in the first 34 nodes contains dual-Intel Xeon 2.4GHz CPUs, 2.5GB memory, nVidia
    Geforce FX5800 Ultra graphics with 128MB memory, an on-board Intel gigabit network interface card, and a Terarecon VolumePro 1000 volume rendering board with 1GB memory.
    Each node of the rest 32 nodes contains dual-Intel Xeon 3.6GHz CPUs, 2GB memory,
    160GB hard disk, nVidia Quadro FX4500 graphics (12 with Genlock daughter
    card) and 512MB memory, and an on-board Inter gigabit network interface card.  The Labs also have a 5-node IBM Visualization cluster with a gigabit Ethernet frontend
    network and a 10Gbps InfiniBand backend network.  Each node contains
    dual-Intel Xeon CPUs, 8GB memory, nVidia Quadro FX3400 graphics card with 256MB memory.

    Stony Brook University has received a Shared University Research (SUR) award from IBM, as part of the company's nationwide initiative to foster collaborative research.  The IBM SUR program is designed to promote collaborative research projects and to increase access to, and successful use of, IBM technologies for research and curriculum.  IBM has made more than $70 million in SUR grants over the last three years in projects ranging from exploration of no-demand supply chains to an effort to find biomarkers for organ transplants. 

    The SUR awards also support the advancement of university projects by connecting top researchers in academia with IBM researchers, representatives from product development, and solution provider communities.  IBM supports more than 50 SUR awards each year worldwide, and including about 10 each year in the U.S.

    As part of the $750,000 SUR award, the Center for Visual Computing at Stony Brook University received a 5-node IBM Deep Computing Visualization cluster, which contains five IBM IntelliStation Z Pro Z20 workstations with a Gigabit Ethernet frontend network and each node is connected to Topspin 120 InfiniBand Server switch (10Gbps) through Topspin InfiniBand Host Channel Adapter.  Each node contains dual-Intel EM64T 3.6GHz/800MHz CPUs, 9GB memory, 73GB 10K SCSI hard disk, a NVIDIA Quadro FX3400 PCI-Express graphics card, an integrated Gigabit Ethernet network interface card, a Topspin InfiniBand Host Channel Adapter.   Each node has 64-bit RedHat Linux WS3 installed.  All the nodes are mounted on an IBM NetBay42 Rack Cabinet equipped with a MRV LX32 Terminal Server, an APC Master Switch, a NetBay local consol manager, a 1U Flat Panel Monitor with a travel keyboard, DPI PDUs.  Projects that will be utilizing the IBM Deep Computing Visualization cluster in the immedediate future include visualization of the Visible Korean Human data set, distributed volumetric ray tracing, and particle flow simulations.

     

    Visualization Cluster

    We are developing a universal processing and rendering system, which utilizes a heterogeneous mix of hardware on a multi-platform cluster in order to process and render very large, dynamic scenes in real-time. Research areas include amorphous phenomena visualization, ray tracing, and massive volume rendering. We are inaugural members of the Hewlett-Packard Scientific Visualization Collaboration Nine nodes are configured as a HP Market Development System (MDS) Visualization cluster. Current applications include rendering of massive volumes which include the full Visible Human data set, several teeth and fossil data sets from the Stony Brook University Anthropology Department, and the Urban Security Project.

    The MDS Visualization system is based on the prototype Sepia-2a hardware compositing architecture which helps overcome the traditional compositing bottleneck in distributed architectures. The MDS system nodes are connected with a ServerNetII high speed switching network. One node serves as a front-end node to this 9-node cluster. The VSS system software is built on NPACI Rocks.   

    Our research includes algorithms for improving the volume memory bandwidth bottleneck, and communication and composition of partial computations and partial images. We are comparing a variety of volume subdivision schemes for memory cache performance. These include block, slice and skewed block volumes. This system is being implemented on top of OpenVL - The Open Volume Library. OpenVL provides an API which abstracts the volume access from the underlying storage mechanism. 

    GPU Cluster for General Purpose Computation

    Inspired by the attractive Flops/dollar ratio and the incredible growth in the speed of modern graphics processing units (GPUs), we propose to use a cluster of GPUs for high performance scientific computing. Our current GPU cluster has 32 nodes connected with Gigabit Ethernet. We have developed a parallel flow simulation using the lattice Boltzmann model (LBM) on our GPU cluster and have simulated the dispersion of airborne contaminants in the Times Square area of New York City. See more information here.