Flux Computation Cluster
To support the computational discovery required for this grant, we will use computing cycles on Flux, a commodity HPC cluster based on the Intel Nehalem platform interconnected with Infiniband networking. Each compute node comprises two 6-core CPUsand 48 GB of RAM (approximately 4GB of RAM per CPU core). The Infiniband interconnect provides 40 GB/s bandwidth and very low latency. If technology advances between thetime of this proposal and the start period of the grant, the system may be upgraded as Flux hardware and software is refreshed regularly to stay current with evolving technology. The Flux cluster includes a comprehensive software suite of commercial and open source research software, including major software compilers, and many of the common research specific applications such as Mathematica, matlab, R and stata. Funds needed for group-specific software will be acquired by individual Project investigators through research grants. The system also includes a 150 terabyte scratch storage using the Lustre parallel network filesystem. This filesystem is for the explicit purpose to allow researchers to store data on a short term basis to solely to perform calculations, and not for any longterm data storage or archive purposes. The cluster and storage is located in a machine room at the University of Michigan. The machine room is a 10,000ft square foot facility with air conditioning and a conditioned, uninterruptible, and generator-backed power supply with sufficient space, power, and cooling capacity to support the proposed hardware as well as any future expansions of this system.This system is connected to the campus backbone to provide access fromstudent and researcher desktops.
Flux provides researchers with experienced full-time IT professionals from the Center of Advanced Computing who have extensive system administration experience with many different parallel platforms including shared-memory and cluster architectures dating back to 1994. Direct research support is provided by the department IT staff as they can best help guide researchers on the use of Flux for their science needs. Flux computing services are provided through a collaboration of University of Michigan units: The Office of Research Cyber infrastructure (in the Office of the VP of Research and the Provost’s Office), the College of Engineering's Center for Advanced Computing (CAC), and Information and Technology Services, as well as the computing groups of each of the schools and colleges at the University.
Accounts will be granted to all faculty; postdoctoral,graduate, and undergraduate students in the Principal Investigator’s research group as appropriate. The computing jobs on Flux are managed through a combination of the Moab Scheduler and the Portable Batch System (PBS). The proposed funds will be leveraged tobudget computing cycles in a way that allows the team of researchers to allocate them as needed, rather thanthe traditional method of buying hardware, which runs down whether ornot it is used. With the Flux paradigm, the size of the allocation can be changed to meet research needs for that given month, allowing for optimal use of the funds allocated.