Overview
Hardware
Bugle consists of 1024 compute nodes (CN) in a single cabinet. Each CN has two processors which makes a total of 2048 processors. The theoretical peak performance of the system is 5.7 TFlops. Bugle has an I/O rich configuration with 128 I/O nodes.
Networks
Bugle has five communication networks:
- A three-dimensional torus for nearest-neighbor calculations on grids (CN nodes only)
- A global tree network for broadcast and reduction operations (all nodes)
- A barrier network for synchronization (CN nodes only)
- Gigabit Ethernet to the JTAG network (all nodes)
- Gigabit Ethernet to front-end nodes (I/O nodes only)
Filesystems
The PVFS2 filesystem is currently used for the /data/ folders which are exported to the Bugle frontend and the Bugle CNs. The fileservers are in RAID 5 configuration.
Summary Table
| Architecture |
2048 processors; 1024 dual PowerPC 440 700MHz 512MB nodes |
| Frontend/Access Node |
4-way Power4+ 1.5GHz (eserver) |
| Service Node |
4-way Power4+ 1.5GHz (eserver), with DB2 UDB |
| Compute Node |
1024 Dual Processor nodes, running CNK |
| I/O Node |
128 Dual Processor nodes, running ZeptoOS |
| Networks |
IBM BGL Torus, Global Tree, Global Interrupt |
| Filesystem |
PVFS2 |
| Disks |
6x5TB storage on 24 disks in RAID 5 configuration |
| Operating System |
ZeptoOS
CNK
SUSE SLES9 Linux
|
| Batch System |
Cobalt |
| Compilers |
IBM C/C++/Fortran
GNU C/C++/Fortran |
| Libraries |
ESSL, MASS |
|
|
.mit.edu
