distem-expe

Script for deploying Distem HPC validation in Grid'5000.

Several ruby scripts aimed at evaluating Linux containers for HPC. The scrips help to deploy and execute NAS parallel benchmarks on:

1. real cluster 2) a cluster made of containers.

How to repeat the experiments

Several parameter for the experiments can be customized using the expe_metadata.yaml file.

log_file: # name of the file to be used as log file.
distem_bootstrap_path: # help script to install and start distem on the machines.
site: # Grid'5000 site
cluster: # specific cluster in the site
runs: #integer which specifies the number of times a measure will be repeated.
timeout: # deprecated
multi_machine: # active the running of scalability test
performance_check: # active the check of performance
kernel_version: # kernel versions to test. The different kadeploy config should be exist
container_tests: # an array for the number of container that will be created on a physical machine
bench_container_test: ?
container_cores: #specifies the number of cores a vnode will be configured with
benchs: # only NAS benchmarks, specifies the type of benchmark and the class to be run.

Linux kernel version and oversubscription

The experiments are controlled by two files: run_all_kernelversion_expe.rb and expe_metadata.yaml. First, you have to download them this repository:

	$ wget https://raw.githubusercontent.com/camilo1729/distem-expe/master/expe_metadata.yaml
	$ wget https://raw.githubusercontent.com/camilo1729/distem-expe/master/run_all_kernelversion_expe.rb

Then, we should use the following metadata:

log_file: "Distem_expe.log"
distem_bootstrap_path: "~/Repositories/ruby-cute/examples"
site: "rennes"
cluster: "paravance"
runs: 20
multi_machine: false
performance_check: true
kernel_versions:
- "4.0"
container_tests: # For the experiment we deploy up to 8 container per machine
- 1
- 2
- 4
- 8
bench_real_test: false
container_cores: 0
benchs:
- :type: lu
  :class: B
- :type: cg
  :class: B
- :type: ep
  :class: B
- :type: ft
  :class: B
- :type: is
  :class: C
- :type: mg
  :class: C

Then, execute the ruby script

	$ ruby run_all_kernelversion_expe.rb

Inter-container communication

Download the following scripts:

	$ wget https://raw.githubusercontent.com/camilo1729/distem-expe/master/expe_metadata.yaml
	$ wget https://raw.githubusercontent.com/camilo1729/distem-expe/master/deploy_lxc_expe.rb

and edit the metadata file like this:

---
log_file: "Distem_expe.log"
distem_bootstrap_path: "~/Repositories/ruby-cute/examples"
site: "rennes"
cluster: "paravance"
runs: 20
multi_machine: false
performance_check: true
kernel_versions:
- "4.0"
container_tests: # number of containers to be created in each machine
- 1
- 2
- 4
- 8
bench_container_test:
- 2
- 4
- 8
bench_real_test:
- 2
- 4
- 8
container_cores: 2
benchs:
- :type: lu
  :class: B
- :type: cg
  :class: B
- :type: ep
  :class: B
- :type: ft
  :class: B
- :type: is
  :class: C
- :type: mg
  :class: C

For the test just one machine will be used. Run the script using the follwing parameters:

	$ ruby deploy_lxc_expe.rb 1 2

Multinode inter-container communication

Download the following scripts:

	$ wget https://raw.githubusercontent.com/camilo1729/distem-expe/master/expe_metadata.yaml
	$ wget https://raw.githubusercontent.com/camilo1729/distem-expe/master/deploy_lxc_expe.rb

and edit the metadata file like this:

log_file: "Distem_expe.log"
distem_bootstrap_path: "~/Repositories/ruby-cute/examples"
site: "rennes"
cluster: "paravance"
runs: 20
multi_machine: true
performance_check: true
kernel_versions:
- "4.0"
container_tests: # number of containers to be created in each machine
- 1
bench_container_test: # depends on the number of machines you reserved
- 1
- 2
- 4
- 8
- 16
- 32
- 64
bench_real_test: # depends on the number of machines you reserved
- 1
- 2
- 4
- 8
- 16
- 32
- 64
container_cores: 16 # depends on the number of cores you have in the machine
benchs:
- :type: lu
  :class: B
- :type: cg
  :class: B
- :type: ep
  :class: B
- :type: ft
  :class: B
- :type: is
  :class: C
- :type: mg
  :class: C

By default it will make a reservation in paravance cluster, you can edit it to change the parameter of reservation.