FY02 I/O Integration Blueprint

Cupps, K ; Gary, M ; Fitzgerald, K ; Quinn, T ; Wiltzius, D ; Minuzzo, K ; Seager, M ; McLarty, T

Lawrence Livermore National Laboratory

关键词: Expenditures;    Testing;    Capacity;    Manpower;    99 General And Miscellaneous//Mathematics, Computing, And Information Science;   

DOI  :  10.2172/15007418
RP-ID  :  UCRL-ID-146419
RP-ID  :  W-7405-ENG-48
RP-ID  :  15007418

美国|英语

来源: UNT Digital Library

【 摘 要 】

This document describes I/O focused requirements, issues, options, plans, deliverables and budgets for Livermore Computing (LC) in FY02. Areas covered include I/O for archival storage, network, platform, visualization and the I/O Testbed. Implementation Plan (IP) milestones and tasks in each of these areas map to the efforts and plans described in this document. When developing FY02 I/O requirements, a survey of key LC customers was performed (see Appendix A and D) and DisCom2 requirements were gathered. The LC customer provided throughput and capacity estimates were quite conservative when compared to ASCI curve projections and were history-based rather than being based on hardware capabilities. Because substantial differences exist in the ASCI platform in FY02, required I/O throughput rates were raised appropriately (i.e., by over 200% platform-to-archive). Archive capacity requirements remain fairly stable in FY02 as aggressive FY01 plans and purchases will accommodate most of the volume of data received through FY02. 10 Gigabit Ethernet network infrastructure will begin to get deployed in early FY02. When full implementation becomes cost effective it will allow us to greatly increase bandwidth between computer facilities. In addition pre-production OC-48 Ultrafastlane encryptors will be installed in early FY02 at each of the Tri-Labs. Preliminary testing of these encryption units and production unit installation will also take place in FY02. Significant network tuning, routing and protocol efforts will be investigated during FY02. These include channel bonding, gigabit ethernet flow control and HPSS protocol enhancements. Changes to and investigation of archive mover platforms and tape technologies are planned in FY02. 1BM tape robotics and tape devices will be phased out. New Storage Area Network (SAN) and platform-based mover architectures will be prototyped and investigated as approaches capable of significantly reducing archive expenditures in support of future 60 or 100 TFLOP platforms. Disaster recovery and RAIT device strategies will be formulated and HTAR porting (machine-to-machine and Linux) and deployment efforts will be completed. The I/O Testbed will continue to be instrumental in pursuit of procuring reliable high performance hardware and software solutions. The Testbed will be used to provide a proof-of-concept for both the direct-to-tape and HPSS SAN investigations, as well as for benchmarking new node and disk platforms and storage devices. Implementation of the FY02 I/O Blueprint will require $2.2M in network investments, $6.25M in archive improvements and a $.37M I/O Testbed investment. Manpower costs will remain constant. The requirements justifying these costs are found in Appendix A. The procurements are detailed in Appendix E.