Brown's new supercomputerDate: Dec 09, 2009
In Providence, R.I., Brown University's Center for Computation and Visualization (CCV) is the new home of a multimillion-dollar IBM Supercomputer that researchers will use to tackle challenges ranging from weather data analysis to predict climate change to human genome mapping.
The new supercomputer runs a total of 1,440 microprocessors and can perform more than 14 trillion calculations per second. It is based on three IBM iDataPlex systems, about the size of six refrigerators; an IBM Cluster 1350; and multiple IBM storage systems running General Parallel File System, supported by IBM Global Services.
Brown also runs a about 200 gigaflops of AMD-based Sun Microsystems rack servers, about 100 gigaflops worth of Dell servers and a 40 terabyte Lustre computer cluster. The college is also installing a small GPU cluster , said Sam Fulcomer, assistant director of Brown's CCV.
To cool the dense computing system efficiently, Brown recently installed a water-chiller system and uses APC in row cooling atop legacy raised floors.
In this video, Sam Fulcomer, the assistant director of CCV, demonstrates Brown's new system.
Read the full transcript from this video below:
Brown's new supercomputer
Sam Fulcomer: I'm Sam Fulcomer, associate director at Brown
Center for Computation
and Visualization. This is our new IBM I-dataplex high performance
computing cluster. Our initial system has a peak performance of over 14
teraflops, discless nodes for high reliability and quad data rate
infiniband as it interconnects and for all [IO] to and from our GPFS file
system which is contained in racks on the other side of the room. We have
over 400 terabytes of raid disc with fully redundant twin tail hardware
raid controllers and redundant GPFS data access servers. We get
approximately five gigabytes a second, with aggregate data bandwidth across
all 12 of the GPFS servers. All GPFS nodes are connected via ten gigabit,
ethernet and quad data rate infiniband. We can provide data services via
CIFS and NFS over the ten gigabit ethernet interfaces to the rest of the
Interviewer: And so, what are you gonna be using these systems, for
Sam Fulcomer: We're using them for computational science applications in a wide
range of disciplines. Bio-informatics, computational fluid dynamics,
computational electromagnetics, computer vision image processing and many
other image processing applications.
Interviewer: And so, what did you have before you installed this system?
Sam Fulcomer: We had a number of very small clusters, primarily with gigabit
ethernet to mirrornet and one small cluster with a ten gigabit infiniband
interconnect. The new system is between one and two orders of magnitude
more powerful than all of its subsystems. So we have more than ten times
the disc space, more than ten times the disc IO bandwidth rates, between 30
and 40 to almost 50 times the peak processing capability than any dataplex.