| |
|
|
|
|
IBM pSeries 690 Technical Summary |
Production date: March 15, 2003
[cu.ncsa.uiuc.edu]
[login-cu.ncsa.teragrid.org]
| COMPONENT |
DESCRIPTION |
Hardware and Networking |
| Model |
IBM pSeries 690 |
| Systems |
Interactive system (cu)
- 32 processors, 64 GB memory
Batch - 11 32-processor nodes
- 7 nodes with 64 GB memory
- 4 nodes with 256 GB memory
|
| Processor type |
IBM Power4 |
|
| Processor speed |
1.3 GHz (5.2 GFLOPS) |
| Cache |
- L1 instruction: 64KB
- L1 data: 32KB
- L2: 1440 KB (shared by 2 processors)
- L3: 128 MB (shared by 8 processors)
|
| Shared gpfs Scratch Disk |
20 TB |
| Interconnect between systems |
Gigabit Ethernet |
| Supported Software |
Operating System |
AIX 5.3 (based on Unix System V) |
| Compilers |
Fortran (77, 90, 95), C/C++ |
| Programming Models |
MPI, OpenMP |
| Program Analysis Tools |
prof, gprof, tprof, Xprofiler |
| Floating point format |
IEEE |
| Batch System |
LoadLeveler |
| Policies/User Limits |
Home directory disk quota |
2 GB |
| Maximum processes and memory per job |
Timeshared:
- 16 processes/threads
- 128 GB memory
Dedicated:
- 32 processes/threads
- 256 GB memory
|
|
| Maximum disk per job |
No limits currently |
| Charging algorithm |
# SUs = ServiceLevel * Time
where:
ServiceLevel = 1.0 for interactive use and
non-Industrial batch jobs
= 1.25 for Industrial batch job
Timeshared jobs:
Time = Total CPU hrs
Dedicated jobs:
Time = Total Wall Clock Hrs * 32
|
NCSA's IBM pSeries 690 cluster is primarily intended to run
applications of low levels of parallelism (1-32 processors) and is also
targeted at execution of third-party applications supported by the AIX
operating system. In addition, four of the 32-processor systems have
256 GB of memory to support large memory applications. Since the cluster does not support multi-node applications using multiple hosts, those who are using
applications requiring use of greater than 32 processors are encouraged to
make use of the Linux clusters or SGI Altix at NCSA.
|
|
|
|
|
