NERSC logo National Energy Research Scientific Computing Center
  A DOE Office of Science User Facility
  at Lawrence Berkeley National Laboratory
 

Franklin for NERSC IBM System Users

This page provides a brief introduction to Franklin for users of the NERSC IBM system Bassi. Please see the other Franklin pages for more complete information.

CharacteristicBassiFranklin
Processor/core type 1.9 GHz Power-5 2.6 GHz Opteron
Processor/core theoretical peak 7.6 GFlop/sec 5.2 GFlop/sec
System theoretical peak (computational nodes) 6.7 TFlop/sec 101.5 TFlop/sec
Number of compute nodes 111 9,660
Processors/cores per node 8 2
Number of compute processors/cores 888 19,320
Physical memory per node 32 GBytes 4 GBytes
Usable memory per node ~26 GBytes (large page) 3.75 GBytes
Number of spare compute nodes
(In service when possible.)
3 20
Number of login nodes 2 16
Node interconnect IBM Federation SeaStar2
File Systems GPFS Lustre
Usable disk space 100 TBytes 350 TBytes
Batch System LoadLeveler Torque/Moab

SSH Access to Franklin

Franklin, like Bassi, uses a particular ssh authentication method called keyboard interactive. End user ssh software must support this authentication method. Other common authentication methods are public key and password.

Passwords and Shells

Franklin uses NIM passwords, which are also used for Bassi, PDSF, Davinci and NIM. Passwords are not changed directly on Franklin, but rather on the Web through the NIM web account interface. NIM is also used to change the default login shell on Franklin.

Charging

Charging is in units called MPP hours, based historically on the IBM Power3 architecture. One Franklin hour is equal to 6.5 MPP hours. See Accounts and Charging on Franklin.

Compiling and Linking

The default compilers on Franklin are Portland Group Compilers (PGI). GNU compilers are also available. The default compilers on Bassi are IBM XL Fortran and IBM C/C++. This table shows a comparison for most commonly used IBM and XT4 Fortran compiler options.

The Cray provided compiler wrappers (ftn, cc and CC) should be used for serial and parallel jobs on Franklin which will be run on compute nodes. The native compilers (such as pgf90, pgcc, gcc, mpif90, mpicc, etc) are not recommended for use on Franklin due to the complexity of the environment.

The default environment on Franklin is 64-bit addresses. The default is also 64-bit on Bassi.

Please see Programming on Franklin for more information.

Running Jobs

Franklin uses Torque/Moab for batch job schedulers. Bassi uses IBM LoadLeveler. Batch jobs should run in user's $SCRATCH directory (instead of $HOME) for better I/O performance on Franklin. Please see Running jobs on Franklin for more information.

File Systems

Franklin users have a $HOME directory and a $SCRATCH directory. The /projects center-wide file system is also mounted on Franklin. Both $HOME and $SCRATCH on Franklin are Lustre File System; on Bassi, $HOME and $SCRATCH are GPFS file systems.

$HOME directory names and default file groups are the user personal file group, which is the same as login name.

Reading Binary Data from Bassi

Bassi is a "big endian" machine, and Franklin is a "little endian" machine. Little endian means the lower-order byte of a number is stored at the lower address in memory while the higher-order byte stored in higher address. It is reversed for a big endian machine. The PGI fortran compiler option "-Mbyteswapio" on Franklin allows codes to read big endian binary files from Bassi. There is no C/C++ option for converting between big endian and little endian binary files.


LBNL Home
Page last modified: Tue, 08 Apr 2008 23:23:59 GMT
Page URL: http://www.nersc.gov/nusers/systems/franklin/for_ibm_users.php
Web contact: webmaster@nersc.gov
Computing questions: consult@nersc.gov

Privacy and Security Notice
DOE Office of Science