Garrett Mc. Wildcarde

Slurm Getting Started / How-Tos

This document is meant to be provide some quick reference materials for gettings started with the basics of working with scripts against a slurm based computational cluster.

The 5 scripts below should all be created or downloaded into the same folder and submit commands run from that base directory.

Notes for use of parallel pcap and auto termination of job on isilon

On Isilon nodes:

Run capstart.sh command to begin rotating parallel tcpdumps on all nodes with external interfaces.

On computation node:

the scripts check-nfs.sh and check-nfs-datacapture.sh must be loaded onto the node and stored in /root to be used via cron.
configure an ssh key on the node that can reach a specific node on the storage cluster as root (required to terminate the server tcpdump process by calling capstop.sh)
Make sure the file /tmp/nfsissuecaptured does not exist
run echo 0 > /tmp/nfsmonitor.semaphore to reset the event notification from previous captures

This works in a python 3.6 environment with bokeh and flask installed. To use this you need the following directory structure:

app/
 - templates/
   - hello.html
 - bokeh-slider.py
 - hello.py

After that you need to open two terminals at app/, in the first one you need to run the bokeh server with the command:

Spock Crash Course

Basic Architecture Info:

~60 nodes
mixture of: ** 24 cores X 250GB memory / node ** 32 cores x 252GB memory / node
No Swap
Dedicated 10Gbps connection to Bucket / node
IB Backend for MPI - not available on 'redshirt' machines

Basic git intro

A few notes, all commands in this tutorial will assume you are working in a linux shell. The same commands should work for OSX but have not been tested. This primer is not intended to give you a full breadth of coverage of the git toolset but show you the basics of making a repository, recording changes, reviewing differences over time, and pushing and pulling from a remote repository. We won't be working with any of the graphical tools for the time being as they are all very different, but the underlying tech to all of them is the same commands we will be doing on the command line.

To get started we'll need a place to work, so lets create that and populate it with some files. I'll be using the files from https://gist.github.com/Wildcarde/9079ead23d5651d871e0 for this demo but feel free to use other files if you would like to.

mkdir gitclass
cd gitclass
wget https://gist.github.com/Wildcarde/9079ead23d5651d871e0/raw/e68a6db01c3e9786d1a49a1b14064dc30f1e18ac/drawfigure.m
wget htt

	version: '2'
	services:
	ldap:
	image: bitnami/openldap:2.6
	env_file: .env
	networks:
	- default
	- proxy
	ports:
	- 1389:1389

	#!/usr/bin/env python
	"""
	Functionally a rewrite of the CSES tool 'add_chassis_nodes(.sh)' however the approach
	is slightly different. This script will add nodes to cobbler using the absolute minimum
	information available. It automates the cobbler command line instead of using the API as
	the developers have deemed the api to be unsafe in it's present state. It could be adapted to
	do XMLRPC as well but is currently assumed to be run on the cobbler server itself so there is
	no need. Some frameworking is done to make it possible to use this file as a component in a bigger
	library, but as of now this approach to using the code is untested.

	function [] = parbench()
	%main function for building pool and running benchmark
	% this is adapted from the mathworks parallel benchmark demo
	% to occupy only one file instead of several
	% original article here: https://www.mathworks.com/help/distcomp/examples/simple-benchmarking-of-parfor-using-blackjack.html

	%% Setup
	%If you want more than 12 core you MUST create the pool by hand
	% Below that it will detect the number of cores requested by and
	% create a pool of the appropriate size. If this pool size is too

	#!/bin/bash

	#This file will create the environment we will be using on spock. A similar environment can be constructed on rondo.
	#First we need to load the module we will be using for this demo, running this command
	#will add a new version of python to your command line. Unfortunately the 3.4 version doesn't have the tools
	#we need, so we will actually be using it to make a custom environment for this demonstration.
	module load anacondapy/3.4

	#Running this will display a list of available python environments you've made with conda
	#we include two stock environments via the module load command and you can make any number

	FROM centos:6
	MAINTAINER Garrett McGrath <gmcgrath815+docker at gmail.com>

	RUN yum install -y wget unzip; /usr/bin/wget "<unreal linux source link here>" -O temp.zip; /usr/bin/unzip temp.zip; rm temp.zip

	RUN yum install -y Xvfb xorg-X11-server-Xvfb which

	RUN chmod +x /LinuxServer/Engine/Binaries/Linux/*

	#this is where you SHOULD be able to mount the confs file and have it 'just work' that's not the case however.