Total storage space¶
Each hub has one disk that stores all persistent user data: home directories and directories for sharing data between users. This disk starts small and grows as needed.
When the disk has less than 10% of its space left, 2i2c’s team gets an alert and increases its size. We may also contact you, in case you want to ask your users to clean up some space.[1]
Usage quotas¶
All of our hubs have a default value of 10GB storage quota per-user, although this may vary depending on the hub.
Also, the shared, and shared-public directories also abide the same default 10GB storage quota.
But keep in mind that the /home/jovyan space is intended only for notebooks and code and is not an appropriate place to store datasets, as it can get really expensive (and slow) when used that way.
Dedicated /tmp disks¶
By default, each user’s /tmp directory shares a disk with other users on the same machine (usually around 20-30GB per user).
If some of your users need more temporary space, you can offer them a dedicated /tmp disk that they pick when they start their server.
Consider this when a few users run workflows that download or generate large intermediate files. Giving those users a dedicated disk is much cheaper than giving every user a bigger machine or a bigger shared disk, because the dedicated disk only costs money while the user’s server is running.
You can offer dedicated disks to all users, or only to some user groups. For example:
To set this up, contact support with the disk size you want and which groups should see it.
Monitoring disk usage¶
You can monitor home directory disk usage for users on your hub to identify large directories and manage storage resources.
Navigate to the Home Directory Usage Dashboard¶
To access the disk usage dashboard:
Navigate to your hub’s Grafana dashboard
Go to Dashboards › JupyterHub Default Dashboards › Home Directory Usage Dashboard
The Home Directory Usage Dashboard displays disk usage for user home directories.
Note that some entries will be for users while others will be shared by all users. Above, we’ve blurred out the users and included the hub-wide directory.
A bigger disk costs more, and shrinking it later is hard. Shrinking means creating a new, smaller disk, moving the data over, and then deleting the old one. So we try to grow the disk only when we need to.