SCIENCE AND ENGINEERING NEWS

The San Diego Supercomputer Center (SDSC) at UCSD has released version 2.0.0 of the popular SDSC Storage Resource Broker (SRB) middleware package, which enables scientists to create, manage, and collaborate with unified “virtual data collections” that are located on heterogeneous data resources distributed across a network. While existing capabilities are preserved for current users, major enhancements “under the hood” give version 2.0 a large number of faster and more powerful services. SDSC SRB version 2.0.0 along with the user manual and release notes are available online.

Interest is growing in the SDSC SRB software because of the need to integrate, manage, and access explosively growing data collections in many fields. Developed by Reagan Moore, Arcot Rajasekar, Michael Wan, and the SRB team in SDSC’s Data and Knowledge Systems (DAKS) program, the SDSC SRB is being used in projects as diverse as helping astronomers integrate multi-terabyte image collections in the NSF’s National Virtual Observatory, enabling NIH-funded neuroscientists to share brain data across the country in the Biomedical Informatics Research Network, developing persistent archives for the National Archives and Records Administration, merging massive sets of NASA satellite data, and bringing together diverse types of environmental data. There are currently more than 200 registered users of the SDSC SRB at more than 50 sites who share the common need to manage, integrate, and collaborate with large data sets.

Beyond the core features of automating many data management functions, planned extensions to the SRB will provide capabilities never before available. “These growing capabilities are enabling new science to be done in new ways,” said Reagan Moore, co-director of SDSC’s DAKS program and lead of the Data- Intensive Computing Environments (DICE) group thrust in the National Partnership for Advanced Computational Infrastructure (NPACI). “Working closely with the computational science community, we have incorporated user requests over the last two years into version 2.0 so that the SRB addresses a growing array of scientific needs.”

New Features in SRB Version 2.0.0 Principal new features in version 2.0 of the SRB (the previous version was 1.1.8) include server-initiated multi-threaded parallel data transfers, which give the new version faster and more robust transfers of very large data sets; revamping the SRB Administration GUI into an easy-to-use Java-based client-side tool that assists in the management of the SRB and the MCAT metadata catalog for such things as creating and deleting users and resources; and parallel bulk loading of metadata into the MCAT, yielding speeds of more than 400 files per second, a factor of 50 faster loading for collections that contain large numbers of small files. Version 2.0 also includes its own Mass Storage System (MSS), which uses a new type of “compound resource” to manage connectivity to tape silos and tape devices, using the SRB to provide caching and other functionality without requiring a proprietary tape management system. The MSS enables users to economically build their own mass storage system in which data migrate automatically between cache and tape.

New features were also added to the MCAT metadata catalog, including access controls on metadata, copying and extraction of metadata, and new attributes for additional capabilities such as locking, hiding, pinning, and versioning. The MCAT has also been ported to work with two new databases, Sybase and Postgres, in addition to Oracle, SQLServer, and DB2.

The installation of the SRB software has been streamlined, and a number of enhancements have made the SRB faster and more usable. In the next few months, the team plans to release Java-based client Application Program Interfaces that can be used to build Web Services Description Language (WSDL) and other web-based interfaces.

“One of the most important new features is the server-driven parallel data transfers,” said Arcot Rajasekar, leader of the DAKS SRB development team. “By incorporating automatic parallel data transfers with up to five threads in a way that is transparent to users, the software optimizes and matches the transfer to the network and server export rates, resulting in transfers that are more robust and two or three times faster.” Early tests have already shown transfer rates at 85% of network capability.

“Other advantages of the server-driven design are that the server has better access to MCAT information to plan an optimum transfer strategy, and data transfer is always directly between the resource server and client, with no intermediate server in between,” said Mike Wan, an SRB developer and senior staff scientist in SDSC’s DAKS group. “In addition, there is good integration with the High Performance Storage System (HPSS) for archival storage, and both client and server are multi-threaded, involving less overhead and consuming less resources.” The SRB parallel transfer uses the more efficient HPSS “mover” API to access data stored in HPSS.

Other new features of the SRB version 2.0 include the capability to synchronize replicated data to ensure accurate mirroring; enhancement to handle the 64-bit architectures of Linux and Solaris; automatic container management, with the new containers created as needed so users don’t need to worry about containers getting full; and reliable file transfers using a persistent transfer mode that automatically retries transfers as needed.

SDSC SRB Version 2.0.0 is supported on the following platforms: UNIX, including Linux Redhat 7.3; Solaris; AIX; SGI; and Macintosh OSX; and Microsoft Windows 2000.

The SRB Data Management Middleware In general, the SRB client-server middleware offers many advantages over, and solves many problems associated with, traditional file systems. What appears as a single collection to the user is in fact a virtual collection consisting of digital entities scattered across distributed, heterogeneous storage resources, including file systems, archives, and databases. The SRB makes all these differences transparent to users, negotiating all protocols, access permissions, etc. across the multiple sites, so that users can access data based on familiar, user-defined attributes and are freed from having to keep track of such complexities as file names, physical locations, protocols, and security arrangements. The capabilities of the SRB support more efficient science at the researcher level as well as enabling new collaborations never before possible.

The SRB organizes metadata about the files in the MCAT metadata catalog to help researchers assemble, search, access, and manage collections of data. The MCAT provides a global name space that spans all the separate resources. Because the MCAT is implemented with relational database technology, it can be extended to include capabilities beyond those of traditional file systems to provide more complex access control systems, proxy operations for such things as delivering subsets of a collection, and knowledge discovery based on system- and application-level metadata.

SRB collections are highly scalable, both in size and in distribution across remote sites. SRB collections at SDSC support over 6.5 million files and forty terabytes of data. Once a collection is created, it can be transparently replicated, managed, and controlled across geographically distributed locations through any of several interactive interfaces: a command-line interface, and new graphical user interfaces including a Windows-Explorer-like interface called inQ — short for inQuisitor — and the Web interface,

Courtesy of 2003 Online, http://www.npaci.edu/online/

============================================================