Name: cloud-benchmarks
Owner: EMBL-EBI Technology & Science Integration
Description: EMBL-EBI test suite for benchmarking cloud providers.
Created: 2015-10-23 09:43:27.0
Updated: 2016-11-29 11:12:42.0
Pushed: 2017-11-20 15:24:14.0
Size: 51
Language: Shell
GitHub Committers
| User | Most Recent Commit | # Commits |
|---|
Other Committers
| User | Most Recent Commit | # Commits |
|---|
The EMBL-EBI Cloud benchmarking script is used by EMBL-EBI to establish a number of metrics that can be used to assess a cloud provider for running EMBL-EBI use cases.
The script is made by a stand-alone bash script that leverages several tests to understand the performances of the underlying infrastructure. At the time of writing, the following tests are included:
All the tools are automatically installed and executed by the benchmarking script, which assumes (and verifies) to be executed in a CentOS 7 environment. The user running it must have sudo rights and its password will be asked to escalate privileges, unless the user has also NOPASSWD rights.
No other OS is currently supported and failure to comply may alter or block tests execution. It is recommended to update the system to the latest packages via yum update prior to execution.
Network connectivity of the Cloud Provider towards EMBL-EBI data centers is tested via GridFTP transfers. To achieve this, some form of authentication is needed. This benchmarking script adopts GridFTP lite to overcome the limitations given by the certificates needed to empower full-blown GridFTP servers, as it exploits SSH to carry out authentication. To allow EBI staff to properly authorize in advance the connection, prospect users must provide the public part of a SSH keypair to be used in the SSH authentication mechanism. Hostname and port to be used for the connection, as well as the assigned username, will be provided by EBI staff as soon as the public key is be uploaded to the server and the relevant user/permission established. Ports from 50000 to 50100 must accept TCP traffic from the outside to allow for GridFTP connections. The benchmarking process require 20GB of available space in the root partition of the virtual machine running it.
Important note
The EMBL-EBI benchmarking server requires authentication. This is currently done via SSH, so you must get in touch with EMBL-EBI providing the public part of the SSH keypair you would like to use to authenticate. EMBL-EBI will then send you back all the configuration details you need to fill the options mentioned below.
Also, keep in mind that you won't be allowed to login into the server anyway, as SSH is only used to authenticate against the GridFTP server. An easy way to test if authentication works is trying to do an SSH on the provided port: getting a “Connection closed” means everything worked as expected, while a “Permission denied” message means that authentication failed. If the problem persists, please get back in touch with us.
The script can be freely downloaded from here. It does need a number of options specified at launch time to correctly carry out the test, as follows:
Mandatory options | Description
—- | —-
--cloud <value> | The name of the cloud provider running the test, this is chosen by the cloud provider itself.
It may or may not coincide with the username assigned by EMBL-EBI
--flavor <value> | The flavor of the VM the test is running on.
Please note that, as part of the test, statistics regarding available resources are acquired
--keypair <value> | The absolute path to the cloud provider private key.
This should match the public key provided to EMBL-EBI for authentication
--port <value> | The port of the EMBL-EBI server to connect to during the test.
This information will be provided by EMBL-EBI staff
--server <value> | The hostname of the EMBL-EBI server to o connect to during the test.
This information will be provided by EMBL-EBI staff
--user <value> | The username provided by EMBL-EBI that will be used for SSH authentication
Optional | Description
—- | —-
--call-home | If specified, test results will be sent back to the EMBL-EBI benchmark server
With all the options correctly specified, the full command to launch the script should resemble the following:
./cloud_testing.sh --cloud=<cloud_name> --flavor=<flavor_name> --keypair=</path/to/keypair> \
--port=<server_port_number> --server=<hostname> --user=<username> [--call-home]
When started, the script will proceed installing all the needed components, both from the official CentOS/EPEL repos and from code available on GitHub repositories. Once this preliminary step is completed, the tests will begin. The process can be followed directly in the terminal or using the verbose log file available at the path printed by the script itself at startup (“Complete log of this run is available at:
If the flag --call-home is specified, all the log files and test results will be automatically compressed and sent back to EMBL-EBI for analysis. Test results and logs can locally be found in the $CLOUD_$FLAVOR folder, within the main results folder EBI_cloud_testing.
It is obviously possible to re-run the script several times. However, to provide the same identical environment in each run, the script takes care of erasing compiled software left from the previous iteration at launch. This does not apply to packaged softwares (i.e. GridFTP executables provided by Globus?s repos), which are not reinstalled every time.
It could be useful to install and use screen in order to avoid accidental benchmark interruption, caused by the disconnection of the shell/network:
sudo yum -y install screen
screen -R bench
Execute the benchmark in the new window:
./cloud_testing.sh --cloud=<cloud_name> --flavor=<flavor_name> --keypair=</path/to/keypair> \
--port=<server_port_number> --server=<hostname> --user=<username> [--call-home]
Press:
CONTROL+ a + d
to exit to the screen session, and:
screen -R bench
to restore it.