Page 17 - OIT Progress Report 2014
P. 17
will SUPPORT
Regular Nodes 20 High Memory Nodes
Heart
Backbone
computer 1,104n o d e s
8
1PB
2redundant Lustre
SMU’s Super Computer “ManeFrame”
Lustre parallel file system
metadata servers
totaling
24GB RAM
CPU cores each
8,832 cores
730 Million 5 megabyte files
Brain
each 192GB
R A M pernode
ManeFrame’s brain is the SLURM scheduling system. SLURM is a sophisticated batch scheduler used for large-scale super-com- puter centers. Typically a job on an HPC system can be classified as interactive or batch. An interactive job is when a user is involved in an interactive use, development, or run of an application on a system. A batch job is a job, task, or run that is queued to be run at the next availability of the requested resource and its associated priority of a user. A batch can further be classified as parallel jobs vs. sequential jobs.
17
with
PLUS
with
