Choose A Resource
As a founding partner of the TeraGrid, SDSC provides three compute resources that are available through the TeraGrid allocations process, in addition to expert assistance available through the Teragrid Advanced Support Program, and a variety of data resources for long and short-term storage and publication of data collections on tape, disk or database. SDSC compute platforms are all scalable parallel-processing systems suited for handling data-intensive computations.
Compute Resources
For access to any of SDSC's computer resources, investigators can apply to the TeraGrid Development Allocations Committee (DAC) for "TeraGrid-wide roaming access," which allows investigators to use any TeraGrid compute resource, or apply for medium (MRAC) and large (LRAC) allocations, which allow requests for time on specific TeraGrid resources.
Each machine has its own key strengths and is specifically designed for different classes of computational problems. When applying for time on SDSC machines, it is important to give careful thought to the appropriateness of the requested resource(s).
DataStar
DataStar, an IBM Power4-based system, permits users to run large-scale parallel applications that exceed the capabilities of typical Linux clusters and provides parallel I/O performance supporting the most data-intensive of applications. In the allocations process, components of DataStar are allocated as different machines, although combined configurations can be arranged for very large runs by special request. Investigators can request time on the following systems:
- SDSC IBM DataStar Power4+ (p655 nodes): The primary DataStar partition All nodes run IBM's AIX OS and are connected by a Federation switch.
- SDSC IBM p690 System:This p690 partition of DataStar consists more powerful processors and greater memory. These nodes also run IBM's AIX OS and are connected to the same Federation interconnect. Additional p690 nodes are configured for login and interactive use, high-performance database serving and HPSS data transfers.
DataStar is ideally suited for parallel applications, with shared-memory nodes and the Federation switch's demonstrated inter-node low latency and high bandwidth. In addition, every DataStar node has a high bandwidth connection to the Parallel File System (GPFS). This configuration has facilitated some of the most data-intensive computations conducted to date. A number of scientific applications and many programming libraries and tools are available on DataStar to assist PIs in their research. For specifications please see:
SDSC IA-64 Linux Cluster
Also called the TeraGrid Cluster, this resource consists of IBM cluster nodes with dual Intel Itanium 2 processors connected by Myricom's Myrinet. Each node of SDSC's IA-64 Cluster has a Fibre Channel connection to a multi-TB GPFS, giving the cluster exceptional parallel I/O performance for the most data-intensive computations.
BlueGene
The triple-rack IBM BlueGene system at SDSC was chosen to support data-intensive computing. Each node consists of two PowerPC processors. All compute nodes are connected by two high-speed networks, a 3-D torus for point-to-point message passing and a global tree for collective message passing. All I/O nodes are connected internally to the global tree and externally via gigabit Ethernet. For details, see:
Data Resources
Users with computing allocations automatically obtain access to SDSC's High-Performance Storage System (HPSS) and SAM Quick File System (SAM-QFS) for storing data.
Database, Collections Disk Space, Dual-Site HPSS Archive and Tape Storage
In addition, any eligible researcher can request a data allocation (with or without a compute allocation) from SDSC that permits expanded access to SDSC's Data Central facilities for data collection hosting, database hosting and long-term archiving. These capabilities make it possible, for example, to share the data from computations or specialized data collections with colleagues or to publish data for access by the broader scientific community.
For details on the available hardware and software and the allocation process, see SDSC Data Central.
