|
|
UMIACS fosters innovative interdisciplinary research by supporting
an advanced research-computing infrastructure with services and
resources for:
High Performance Computing on a variety of
architectures including clusters of computers acting as a
distributed-memory parallel systems, Symmetric Multi-Processing
systems, and General-Purpose GPU systems.
Distributed Computing on a high-speed Local Area
Network with load-balancing services and fast Wide Area Network
connectivity through the Mid Atlantic Crossroads (MAX), the Next
Generation Internet Exchange (NGIX), the Internet2, and the National
Lambda Rail.
Data Intensive Computing on a variety of disk,
tape, and file-system platforms that host major data collections for
our labs and the national research community. The Institute currently
maintains over 400TB of long-term persistent data in the areas of
Medical Imaging, Computational Biology, Computational Linguistics,
Computational Fluid Dynamics, and Computer Vision.
Visual Computing on very high-resolution tiled
displays, immersive visualization environments, and 3D stereoscopic
displays.
Private Cloud and Web Hosting environments that
run on the VMware and KVM hypervisors that support Applications,
Platforms, and Infrastructure as services to our labs.
Each of the Institute’s facilities is lead by distinguished faculty
members who direct the work of our researchers, systems
administrators, network engineers, and programmers. These
interactions between academic researchers and information technology
professionals equip the Institute’s leading research programs with
cutting edge technologies and forward-thinking infrastructure.
While no single system is extremely large in scale, the site as a
whole is a large installation with 1200 supported computers, 2500
network ports, and 400 Terabytes of managed data. Some of our
deployments include:
- The Chimera Cluster is a high-performance computing and
visualization cluster that takes advantage of the synergies afforded
by coupling central processing units (CPUs), graphics processing units
(GPUs), displays, and storage under an infrastructure grant from the
National Science Foundation. The infrastructure is being used to
support a broad program of computing research that revolves around
understanding, augmenting, and leveraging the power of heterogeneous
vector computing enabled by GPU co-processors. The cluster is
comprised of: a 128-processor Linux-based visualization cluster built
on the Intel Xeon platform and interconnected with Infiniband. Each
node has twenty-four GB of memory and NVIDIA Tesla S1070 GPU. The
nodes are coupled with an 8 Terabyte shared file system and a
50-megapixel display wall that is made up of twenty-five LCD monitors.
- The Skoll cluster, supported by funding from the Office of Naval
Research (ONR) and DARPA, is dedicated to exploring the possibilities
of Distributed, Continuous Quality Assurance, and is comprised of: a
120-processor cluster running a mix of Operating Systems including
Linux and Windows built on the Intel Pentium platform and coupled with
a 3-terabyte Network Attached Storage System.
- The CBCB Computing Facilities. Scientists at the Center
for Bioinformatics and Computational Biology are involved in many
different genome sequencing projects, both as principal investigators
and as collaborators. CBCB brings together scientists and engineers
from many fields, including computer science, molecular biology,
genomics, genetics, mathematics, statistics, and physics, all of whom
share a common interest in gaining a better understanding of how life
works. The primary computing cluster consists of: a 288-core
Linux-based cluster built on the 64-bit AMD Opteron platform and
interconnected with Gigabit Ethernet. These nodes are coupled with a
thirty terabyte network attached storage system, a Mysql relational
database that provide access to data stored on a 3par Inserv Storage
Server, and a 128 processor Hadoop cluster with 20TB of usable
storage. The cluster also includes several large memory computing
nodes built on the AMD Opteron platform: Four dual-processor nodes
each have eight Gigabytes of memory and three quad-processor nodes
each have thirty-two Gigabytes of memory.
|
