Mass Storage System (MSS) improvement plans
A number of improvements are scheduled for FY2006. Some of them will phase out old technology enabling future growth; others will enhance system tracking and system management. Projections from the SCD Strategic Plan indicate a 25-fold increase in computing capacity (see the figure below). Historically, the amount of data managed by the MSS has grown proportionally with the computing capacity. These improvements will ensure that the NCAR MSS continues to scale and meets the needs of the NCAR scientific community.
|
Hardware improvements
Technology refresh occurs on regular cycles depending on the equipment type. Historically, tape and disk drives have a three to five year life span in the MSS. Data channels have a longer life span. The following technology changes will be completed in FY2006.
HiPPI (High Performance Parallel Interface) is an 800 Megabit per second data interface introduced to the MSS in the early 1990s. Vendor support of HiPPI has waned in recent years as other less costly technologies have appeared. The deployment of the MSS Storage Manager in FY2003 introduced Gigabit Ethernet (GigE) as an alternative data interface. GigE has a long future because it has become a ubiquitous interface, it has commodity pricing, and its speed will increase over time. The MSS use of HiPPI will end in FY2006 with the decommissioning of the HiPPI storage devices. After this, GigE will be the sole MSS data transport interface.
The original 11-TB MSS Storage Manager disk cache will be replaced in FY2006. This disk, a StorageTek D178 disk system, uses Fibre Channel disks that were once state-of-the-art technology. Unfortunately, this disk system will no longer be supported by the vendor after the end of calendar year 2005 and therefore must be replaced in early FY2006. In the past few years, Serial ATA, or SATA, disk technology has grown in acceptance and deployment and is now an option for replacing more costly Fibre Channel disks depending on data access patterns. A SATA evaluation was completed in FY2005 (see MSS accomplishments) and concluded that SATA technology meets the requirements for the disk cache. Therefore, 44 TB of SATA disk will be deployed in FY2006 to replace and enhance the original disk cache. The expanded disk cache will reduce the number of tape drives required to maintain the required data flow as more of the active data will reside on the disk cache, resulting in improved data delivery response times.
Additional StorageTek 9940B tape drives will be deployed in FY2006. The 9940B supports a 200-GB tape cartridge, a 3.3 times increase over the 9940A format using the same cartridge. The additional 9940B drives will enable the migration of existing data to the higher density media, thus increasing the total capacity of the existing tape silos to a total of ~6,000 TB or 6 petabytes (PB). Reuse of the existing tape cartridges will provide a significant media cost savings. As data are copied from the 9940A format, the empty tape cartridges are reused in the 9940B format. This reuse will minimize the amount of tape media purchases over the next few fiscal years. Currently data are physically separated between manual-mount shelf storage and robotic silo storage. The manual mount shelf storage data are being migrated to robotic silo storage as they are rewritten in 9940B format. Support for manual-mount tape drives will no longer be needed once this migration is complete.
Software improvements
The metadata server project will be completed in FY2006. This project will replace the existing metadata database running on the MSCP with a commercial IBM DB2 database running on a Unix server. The DB2 database will provide higher scaling, reduce the dependency on locally written software, and enable statistics and tracking metrics previously unavailable.
The Distributed Computing Services (DCS) software provides the user interface to the MSS. It is a locally written interface that relies on the Open Systems Foundation (OSF) Distributed Computed Environment (DCE) software. The use of DCE is declining, as is its support on current operating systems. To improve the portability of DCS, a new version is in development. DCS version 4.0 is being rewritten to remove the dependency on the OSF DCE library, with the goal of being more easily portable. The first deployment of DCS V4 will be in early FY2006 on NCAR's IBM BlueGene/L research machine. Deployment of DCS V4 will continue in FY2006.
Business continuity
A project was started in FY2005 to address data recovery under the context of a disaster. A requirements survey was completed as the first phase of the plan. The second phase, investigating options, is in progress. Possible options range from manually transporting and storing data tapes at one of the other UCAR campuses, to installing a robotic tape library at one of the other UCAR campuses, to expanding the MSS to be fully redundant at a second site. Subsequent steps will be addressed in FY2006. Initially a subset of the MSS archive will be targeted for data recovery. Most likely this will be a subset of the Data Support Section scientific data archive. Including other user data will be investigated as the project progresses.
 |
FY2005 Annual Report |
|
|
|