Fedora Server Installation Guide

A new Fedora installation creates a (modern) GPT partition table by default.

On a BIOSboot machine, Anaconda creates at first a small (1 MiB) BIOS boot system partition on the first drive. It stores the second stage bootloader which is required by GNU/Grub. Subsequently, it creates a /boot` partition of 1 GiB. It contains all the files necessary for booting Linux, especially the kernel. The remaining area is completely filled with a third partition containing one large volume group (LVM VG) named fedora by default. You end up with 3 primary partitions on the hard disk that use all the available space.

Fedora can still use the (legacy) MBR partition scheme, provided that the disc is not larger than 2 TB. It then omits the `BIOSboot partition and uses only the other two partitions.

In the case of a UEFI boot system, Anaconda creates first the required 'EFI System' partition and then adds the aforementioned /boot partition and one large LVM partition and Volume Group (VG) as described above. You will end up with 3 partitions on the hard disk that completely occupy the available space.

In each of these alternatives, Anaconda creates one logical volume of approximately 15 GiB (the exact value depends on the disk capacity of your system) named root for the operating system and its software. The remaining available space is at the disposal of the system administrator for free use to store user data.

The rationale behind this is a separation of system and user data, which eases system administration, increases security, and decreases the likelihood of errors and data loss. The system area (i.e. the operating system including installed software) must be maintainable completely independently of the storage of user data. System maintenance must not jeopardize user data under any circumstances. If necessary, it must be possible to unmount user data.

Following this principle, the system administrator would later set up additional logical volumes for storing an application’s data and mount them at an appropriate location in the directory tree. In case of a PostgreSQL database, for example, a system administrator would create a logical volume of appropriate size, assign a descriptive name, such as pgdata, and mount it in the directory tree at /var/lib/pgsql, where Fedora PostgreSQL expects the data to reside.

In this way, any error that may occur in the file system should have as little impact as possible and jeopardize as little valuable data as at all possible. For this, the additional effort in system administration is purposely accepted.

If you are a more experienced administrator, you may wish to further the rationale above with increased separation.

You will select Custom and create the BIOSboot, efi and /boot partitions as required and a small partition and VG dedicated to the operating system. A good size for this VG (eg. sysvg) is, approximately, 30 GiB. Occupying the remaining space, you will create a dedicated partition and Volume group (eg. usrvg) for user data. You will end up with 4 partitions on the hard disk (boot, sysvg, usrvg with Bios boot machines and hard disks up to 2 TB) rsp. 4 partitions (BIOSboot/efi, boot, sysvg, usrvg for all other machines) that use all the available space.

Create a LV (e.g. sys_root) of about 15 GiB for the operating system and maybe additional LVs for the runtime environment, e.g. a LV sys_log of about 5 GB. Mount it at /var/log to prevent log files from flooding and blocking the system and, vice versa, prevent that any other space issue on the root partition block your logs and complicate error analysis. The remaining free space is left for distribution as needed over time. Similar to the default partitioning, all user data is created as LVs in usrvg and mounted in the corresponding directories of the system. This is the maximum possible separation of system and user data with only one hard disk available. And with today’s typical hard drive size of 2 TB and more, those dedicated 30 GBs don’t interfere with the effective use of disk space anymore.

If there is more than one disk available, the default partitioning creates, on each of the other disks, one big partition with a Physical Volume (PV) and adds it to the VG.

On a server, this is usually not optimal. Rather, several disks should store data redundantly in order to maintain operation in the event of a hardware failure. Configuring a RAID system is one such solution.