Skip to content

Files and Dirs

UI defaults

This YAML file is used as input for both CLI and GUI to deploy environments, collect data, plot graphs, and get advice.

It currently has these fields:

  • subscription: subscription name used to provision resources
  • rgprefix: prefix of resource group name used to host all resources
  • region: Azure geographical region of deployment (e.g. eastus, southcentralus,...)
  • appsetupurl: url that contains the bash script with instructions on how to download input data and application, and how to run the application
  • skus: Azure SKUs (VM types) to be tested
  • nnodes: Number of nodes to be tested
  • ppr: percentage of processes per resource
  • appinputs: application inputs (e.g. matrix size and number of execution interactions for the matrix multiplication application)
  • vpnrg (optional): existing resource group that contains a vpn setup
  • vpnvnet (optional): existing vnet name for the vpn setup
  • peervpn (optional): boolean for peering with vpn resource group / vnet
  • createjumpbox (optional): boolean for creating a VM in the same resource group

Example inside matrix multiplication folder:

ui_defaults.yaml
subscription: mysubscription
skus: [Standard_HC44rs, Standard_HB120rs_v2, Standard_HB120rs_v3]
rgprefix: nettoaha
appsetupurl: https://raw.githubusercontent.com/Azure/hpcadvisor/main/examples/matrixmult/appsetup_matrix.sh
nnodes: [2, 3, 4]
appname: matrixmult
tags:
  appname: matrixmult
  version: v1
region: southcentralus
createjumpbox: true
ppr: 100
appinputs:
  appinteractions: 3
  appmatrixsize: [4000, 7000]

Data filter

This YAML file specifies which parts of the dataset one wants to use for plotting graphs or getting advice (pareto-front). You can specify the application name, and the deployment environments, which can be useful in case data was collected using a few deployments but one does not want to consider all deployments for that particular application.

Example inside matrix multiplication folder:

datafilter_matrixmult.yaml
appname: matrixmult

# uncomment to filter more

#deployment: mydeployment240712v1
#tags:
#  version: v1

Main HPCAdvisor dir

All files related to deployment, dataset, task output, among others are stored in the main HPCAdvisor folder, located at:

$HOME/.hpcadvisor/

Deployment dirs

Every deployment contains a directory located at:

$HOME/.hpcadvisor/<deploymentname>

Task stderr and stdout

For each task executed, two files are collected:

Standard error and standard output:

$HOME/.hpcadvisor/<deploymentname>/output-<taskname>/stderr.txt
$HOME/.hpcadvisor/<deploymentname>/output-<taskname>/stdout.txt