Sun Fire 6800
The primary compute server is a Sun Fire 6800 powered by twenty 1.2GHz UltraSPARC IV processors and four 1.5GHz UltraSPARC IV+ processors - making a total of 48 logical processors available (from Chip Multi-Threading). It also has 96GB of shared memory, fibre attached Ethernet and is connected to our SAM-QFS data storage environment. At this time, the 6800 has been partitioned into two domains - one with 20 processors (40 logical), 80GB memory and the other with 4 processors (8 logical) and 16GB memory. Resources such as compute and memory can be dynamically allocated between the domains as required.
The 6800 domains available are
The MAGPIE cluster consists of 24 SunFire v210 servers with dual UltraSPARC IIIi processors (12-1GHz nodes, 12-1.3GHz) and 2GB RAM. The cluster is connected to the SAM-QFS file systems through the master node - which handles file services and management. The interconnect is a dedicated 24-port 1GB switch.
MAGPIE Cluster Machines
The software environment consists of the Solaris 10 Operating Environment, Sun HPC Cluster Tools and Sun Studio. The cluster is most often used by MAGPIE however, it can also be used for shared-memory (OpenMP) and message-passing (MPI) applications.
TimeLogic DeCypher Systems
The COE possesses eight TimeLogic DeCypher engines for high-throughput bioinformatics database searches. Using field-programmable gate arrays (FPGA), the DeCypher engines can be reprogrammed on-the-fly making them highly versatile, running both heuristic (shortcut) and canonical methods by simply loading a different hardware image. The DeCypher engines enable highly sensitive sequence similarity searches such as the various types of Hidden Markov Model (HMM) or Smith-Waterman analyses that would otherwise be too large for public resources and too time-consuming for local computer clusters.
The version of BLAST from TimeLogic runs approximately 30-times faster than the original for large data sets. TBLASTX runs 1.5 times faster than on the 24 CPU BlastMachine. Gene-BLAST has also been developed to allow rapid alignment of ESTs to genomic sequence (intron spanning), something BLAST cannot do. Hidden Markov Models of protein families run more than 100 times faster than software methods running on a cluster at Washington University at St. Louis.
Paracel BLAST Machine
This machine consists of a 12 nodes containing dual 2.3GHz Xeon processors, 2GB RAM; all running Linux. Paracel has optimized the hardware configuration, as well as tweaking the BLAST source code itself, to maximize throughput of NCBI BLAST. This is the accelerator used to run methods not ported to the other hardware systems described, such as PSI-BLAST (iterative alignment) and MEGABLAST (genome-to-genome style whole DNA comparisons).