“virtualization” per se, but the “virtual”
designation fits in that users access the
application remotely, not on their local
computers. The project’s most notable
innovation was software-driven management logic that provides resources
as needed and allows unused resources
to be used for HPC applications, including molecular analysis. Students could
use any Web browser with access to adequate bandwidth (at least 125kbs) to
connect to dozens of desktop applications anywhere/anytime.
We intended to deploy this innovation at NCCU using blade servers—
ultra-thin computers with multiple
high-end processors—in a highly flexible “one-stop-shop” infrastructure
to provide the same service to our students and faculty. With the two major
biotechnology centers at NCCU, along
with the university’s focus on scientific computing, we felt the NC State
approach would also be appropriate
for NCCU—use the blades to run vir-tualized applications when needed but
apply all idle processing time to long-running, processor-intensive scientific
applications.
The goal was a scalable, reliable in-
frastructure for both virtualization and
HPC applications. Toward this end, we
began by purchasing nine blade serv-
ers. In fall 2005, we received a hardware
grant from IBM ( www.ibm.com/univer-
sity) providing $84,000 for hardware,
though nothing for soft ware or support.
Most was apportioned to the infrastruc-
ture to support the blades, leaving little
for the blades themselves. For example,
the rack required to host the chassis for
the blades cost $2,649. The network
switch for the blades cost $10,000.
The monitor, keyboard, and video and
monitor connector cost $2,245. After
we ordered these foundational pieces,
there was funding enough for nine ini-
tial blades. We chose IBM HS20 Xeon
blade servers with 4GB of RAM and
two 3.8GHz processors per server. Each
server also had two 36GB mirrored
hard drives. The cost per blade, with ex-
tra processor, memory, and hard drive,
was approximately $6, 106.
