PSIgroup | Computing / Globus

Globus is a data management platform that orchestrates moving large dataset between computing resources. You must have a compatible identity provider, such as orc.id.

Personal

To use as a single user on a machine, Globus Connect Personal can download from and upload to Globus servers, but cannot interact with other Personal nodes (see Server). This is installed on our linux machines at `/opt/globusconnectpersonal`.

To setup globus for your user account, run /opt/globusconnectpersonal/globusconnectpersonal -setup. This will prompt you to log in to Globus in a web browser and return with a authorisation code. Once setup completes, configuration will be stored in ~/.globusonline. The client can be run headless with /opt/globusconnectpersonal/globusconnectpersonal -start but a Systemd user service can manage the client for you:

Create the file ~/.config/systemd/user/globus-personal.service with the following content:

[Unit]
Description=Globus personal transfer daemon

[Service]
Type=simple
ExecStart=/opt/globusconnectpersonal/globusconnectpersonal -start
Restart=on-failure

[Install]
WantedBy=default.target

You can then start the service with systemctl --user start globusconnectpersonal. Replace start with status to check the service status and with stop to shut down the service when you are done. See e.g. the Arch wiki on systemd/User for more on managing user services.

With Globus running, you can set up data transfers at https://app.globus.org/file-manager. You can browse datasets, such as those in the Human Organ Atlas at https://app.globus.org/file-manager?origin_id=340dc883-4b0d-476c-abb0-969e1ddd9dc0.

Server

The Globus Connect Server allows multiple users to access a shared storage resource. Unfortunately it must be hosted on a machine that is publicly addressable from the internet. It also requires a set of ports to be exposed for control (TCP 2223) and data transport (TCP 50000-51000). These ports were opened for our internal subnet, which allows the Globus Personal client to work, but a public IP would need to be exposed to set up a Server.