Press Archive

SDSC Deploys SAM-QFS System Reducing Metadata Backup Time From 21 Hours to Just 30 Minutes

New system provides tremendous input/output capabilities for increased speed and to ensure performance for future growth

Published 11/21/2005

The explosive growth in data intensive computing has led San Diego Supercomputing (SDSC) to deploy SAM-QFS, a high-performance archival storage system for files requiring frequent or dynamic retrieval. As a result, SDSC's backup time has decreased by nearly 4000 percent.

SDSC recently installed and deployed a 32 GB RAM-based solid state disk from Texas Memory Systems for the metadata storage. This 500+ Terabyte system allows users to access data directly using a disk cache file system, which automatically migrates data to tape nightly. Although files on SAM-QFS appear to be on a live-disk cache, they are actually stored on tape. SDSC set a goal of performing metadata backups four or more times a day, providing better data integrity to the archive data.

After extensive evaluation, the Texas Memory Systems RamSan-325 solid state disk system clearly met the needs of SDSC's SAM-QFS Archival system. The RamSan-325's tremendous input/output (I/O) capability ensures that the metadata device will not bind the archive and provides performance capacity for the archive's future growth.

The RamSan-325's performance exceeded SDSC's performance requirements, reducing the metadata backup time by from 21 hrs and 40 minutes to only 34 minutes. This change resulted in a huge decrease in the completion time for file system operations across a range of activities. Given this metadata performance enhancement, SDSC will be able to improve data integrity in the file system by conducting more frequent metadata backups. Finally, archive users will enjoy the quick response to metadata queries from the archive, making their interaction with the archive more efficient and productive.

SAM-QFS utilizes SUN's QFS file system, which allows multiple servers to attach directly to shared disk storage over a storage area network (SAN) and view all of the storage as through a single file system. This approach provides higher data transfer bandwidth to the file system than any one server could provide. However, it requires complex file system management that maps access to files and directories using metadata to decode the physical file location on the shared disk storage. This metadata is a small component of a file system; however, it experiences a disproportionately large amount of I/O than its size would suggest.