The National Center for Supercomputing Applications is rolling out a storage infrastructure that will include 380 petabytes of magnetic tape capacity and 25 petabytes of online disk storage made up by 17,000 SATA drives.
The massive storage infrastructure is designed to support of one of the world's most powerful supercomputers, called Blue Waters. Commissioned by the National Science Foundation (NFS), Blue Waters is expected to have a peak performance of 11.5 petaflops, though the specification given by the NFS is for it to offer 1 petaflop of sustained computing power for applications.
The NCSA, which is run by the University of Illinois, has awarded Cray a contract to build the supercomputer. The system will run a Lustre parallel file system with more than 1 terabyte per second of throughput to its back end storage.
The Blue Waters project will create a 1 petaflop supercomputer to handle real-world science and engineering applications. Among others things, it will aid in understanding how the cosmos evolved after the Big Bang, help predict the courses of hurricanes and tornadoes, and play a role in the design of new materials at the atomic level.
The supercomputer will be made up of more than 235 Cray XE6 cabinets using 380,000 AMD Opteron 6200 Series x82 processors and more than 30 cabinets of a future version of the recently announced Cray XK6 supercomputer with 3,000 NVIDIA GPUs. The system will include 1.5PB of aggregate memory from 190,000 memory DIMMs.
In support of all that computing power, the NCSA is deploying 25PB of disk storage using Cray Sonexion storage systems. Sonexion is a rebranding of Xyratex storage arrays. The system offers up to 1TBps of aggregate bandwidth over a 40Gbps Ethernet from Extreme Networks.
"We've been working heavily with ... networking vendors to make sure they're ready to go with the 40 gigabit Ethernet," said Michelle Butler, the NCSA's senior technical program manager in charge of storage and network engineering. "We're not the first to use 40Gbps Ethernet, but we're among only [a few]."
Key to using the 40Gbit Ethernet network is the ability to carve up the pipe into multiple 10Gbps Ethernet channels, to enable the NCSA to spread the fabric across multiple ports, Butler said. The Ethernet will be used to connect about 75 hosts.
The NCSA also selected a DataDirect Network's SFA 12K storage array delivering 100GBps of storage performance to offload data to the "nearline" tape library system. The disk subsystem is scalable to 500PB of capacity, Butler said.
"That subsystem has to be able to offload that terabyte-per-second file system, so we needed a very large tape drive infrastructure to be able to offload that," she said.