Amethyst Replacement Nodes

To upgrade the Amethyst disk array from 20 TB usable storage to 100 TB usable storage, as well as increase its capability to handle increased user load with 6 Amethyst replacement nodes with Graphstream P2X12B-R1B is a 1RU-rackmount server unit that can support GPU-computing applications more efficiently than currently available alternatives. The Data Analysis and Assessment Center (DAAC) relies upon two main storage arrays. The first storage array is connected to Amethyst, which uses a Linux-based operating system. This 20TB disk array was installed in January 2007 and consistently runs over 80% full. The Amethyst disk array is used by DAAC staff and HPCMP users. Amethyst is no longer able to efficiently process data sets, even with data staging, since current data sets are approaching 50 TB in size. The Amethyst disk array suffers not only from a capacity problem; it also suffers from a capability problem. Soon after installation, it was discovered that when more than six instances, either serial or parallel programs, access the file system, processing on Amethyst slows down considerably. This can easily be accomplished by having one person perform a parallel compilation, “make –j6”, of code. If two people are performing parallel compiles, the system grinds to a halt. Therefore the DAAC needs a system that can handle multiple users with multiple concurrent I/O streams