PresentationonMRIfunding-DCC-Ligo.pptVIP

LSC General Meeting 2000.08.17 LIGO Scientific Collaboration - University of Wisconsin - Milwaukee UWM Data Analysis FacilityLSC General Meeting, August 2000Bruce Allen, Patrick Brady, Alan WisemanUniversity of Wisconsin - Milwaukee History January - March 1998: built 48-node beowulf system for binary inspiral search project. Proved valuable: inspiral search coderequired 100 runs to “get it right”. System useful for users outside UWM Julien Sylvestre (MIT), Matt Evans (Caltech) R. Balasubramanian (Cardiff) Olliver Wehrens (AEI) Offered a UWM MRI slot (matching funds) in December 1999 to upgrade system before LIGO-I science run. Funded by NSF: July 2000. MRI Development Proposal $415k (NSF) + $177k (UWM). Baseline design: 128 nodes, 10 Tbytes disk, mix of channel-bonded 100 Mb/s and 1 Gb/s networking. Part-time technician to help with assembly, routine maintenance. System designed for data analysis prototyping and development work. Will also serve as GriPhyN testbed. NOT a 24x7 reliable system. No staging of data from tape needed. Disk failure/data loss tolerated. Representative data set on disk (hours all channel data, days of reduced data, weeks of very reduced data). LSC Involvement In addition to UWM users, proposed to make system available to other LSC members. Support 9 LSC users, chosen 3 times/year by committee: Allen, Brady, LIGO Lab director, LSC spokesman, software coordinator In early January (~2 weeks before submittal) proposal sent to LIGO Lab Director and LSC Spokesman for comments They wrote a letter of support to the NSF Asked that a committee should review final design before purchasing/procurement phase Design/Procurement Plan Proposal contains a complete baseline design, with detailed costing and quotes. Remainder 2000: benchmark and prototype different CPU, networking and disk options. Make final design Early 2001: assemble small committee of experts to review final design, and make recommendations about any remaining technical