This lesson is being piloted (Beta version)

New User Training: Glossary

Key Points

Introduction to Blue Waters
  • Blue Waters enables Petascale Science not possible anywhere else

  • Key emphasis is on sustained petascale computing

Logging in
  • Blue Waters general address is bw.ncsa.illinois.edu

  • Blue Waters uses two-factor authentication

  • ~/.ssh/config stores important connection information

  • -X flag enables X11 forwarding over ssh

Filesystems
  • Blue Waters has 3 online filesystems and one tape storage system

  • projects command lists user’s science projects on Blue Waters

  • quota command lists user’s usage on all 4 filesystems

  • Blue Waters supports ACL for for fine-grained access control to files

  • setfacl and getfacl

Data Transfer
  • Use scp, rsync, wget, git for “small” data transfers

  • Use “Globus Online” (GO) for large/long data transfers

  • Blue Waters’ and Nearline addresses in GO are ncsa#BlueWaters and ncsa#Nearline, correspondingly

  • Use Globus to store data in Nearline

Software Stacks
  • Software on Blue Waters lives in “modules”

  • “Install” and “uninstall” software with module load / module unload

  • There are 4 types of Programming Environments on Blue Waters: Cray, Gnu, Intel, and PGI

  • Python and its derivatives (Tensorflow, PyTorch) live in “bwpy” module and its sister modules.

Running jobs
  • Job scripts are regular scripts with special #PBS comments

  • #PBS comments must appear at the top of the job script

  • #PBS comments specify all requests and set characteristics of compute jobs

  • Job scripts are executed on MOM nodes.

  • Interactive jobs are the jobs submitted with -I flag and do not accept job script as input

  • aprun must be used to send applications for execution to compute nodes.

Glossary

Blue Waters Portal
Blue Waters main website: https://bluewaters.ncsa.illinois.edu
Compute Node
A node where (part of) your job is executed
Login Node
A node where you land when you connect to Blue Waters using SSH
MOM Node
A node where you land when your job (interactive or batch) starts.
Node
Analogue of a standard computer in a supercomputer.