Mailinglist Archive: opensuse-storage (4 mails)

< Previous Next >
[opensuse-storage] Notes from Storage meeting

now that hopefully everyone is on the list, find attached the notes from
the meeting we had in Nuernberg.

It is based on the Etherpad notes created during the meeting.

After getting some feedback, I would push it back to Git so that it can
be worked on further as the requirements of various areas and details
about some specific stuff becomes clear.

Thanks for any feedback, and sorry for taking that time,


Jiri Srain
Project Manager
SUSE LINUX, s.r.o. e-mail: jsrain@xxxxxxxx
Lihovarska 1060/12 tel: +420 284 084 659
190 00 Praha 9 fax: +420 284 084 001
Czech Republic

# YaST Storage Features and Requirements

This is about *libstorage* and the *yast-storage* Ruby code.

##AIs summary

AI Lukas: ML: libstorage-ng@xxxxxxxxxxxx
AI PM: Clarify requirements on FAT partitions
AI lmb, Holgi: Claryfy requirements on filesystems (see below)
AI Arvin: clarify why: reate partitions on RAID (if there already is a
partition table; cannot create a partition table if there is not one already
AI Architecture maintainers: Clarify which configurations of RAID are bootable,
possibly under which conditions
AI: Veryfy support of resizing logical volumes for LVM
AI: Clarify why we do not support encryption of root filesystem, which tools
would possibly need to be fixed
AI: speak with different parties which requirements they have on automated
proposal (Cloud, SES, SAP, ...)
AI: Define when dedicated boot partition is needed before merging into the
AI: Find otu if EVMS support for upgrade is necessary (when was it last
AI: Provide specifics for zFCP devices
AI: Define requirements on /boot/zipl partition
AI: Provide details about specific tools for first sector writing, reasoning
for two copies of kernel (why two?) for DASD

## Disks and Partitions Simple case:

* Disk
* Partition table
* Partition
* File system
* File
* File
* File
* ...
* Partition
* File system
* File
* ...
* Partition
* ...
* ...

## Advanced Cases

This covers various ways to combine disks and/or use advanced
ways to manage the disk space provided by the disks

###RAID (Redundant Array of Independent Disks)

* Hardware RAID: mostly automatic and transparent
* Behaves like any regular disk
* No need for any special handling (or even detection)
* Linux Software RAID / BIOS RAID (type dependent on BIOS):
* mdadm RAID:
* dmraid (Device Mapper RAID): ?

BIOS-managed raid is configured in BIOS (and Linux system detects it), Linux
softwre RAID must be configured in Linux

RAID levels:

* **RAID 0**: striping; performance improvement due to parallel disk
I/O; less fault tolerance than one disk
* **RAID 1**: mirroring; improved fault tolerance, same performance as
one disk
* **RAID 5**: 3 or more disks with redundancy (for parity blocks): Good
fault tolerance, improved performance (parallel I/O)
* **RAID 6**: 4 or more disks; more redundancy than RAID 5, thus even
better fault tolerance
* **RAID 10**: striping + mirroring; combination of RAID 1 and RAID 0

Supported actions:

* create RAID 0, 1, 5, 6, 10
* create partitions on RAID (if there already is a partition table;
cannot create a partition table if there is not one already- AI: Arvin - why?)

Not supported by YaST storage - use dedicated tools instead:

* add another disk (to already created RAID)
* remove disk (from RAID)
* create BIOS RAID (use BIOS or mdadm/dmraid)
* These cases need to be documented properly

Raid and boot partitions

* AI: Define which Raid configurations are bootable?

### Volume Managers

* Btrfs (volume manager features): not yet supported, needs FATE with
usecases and technical information
* EVMS: legacy; no longer supported, drop
* AI: Find out if we need upgrade support of EVMS

Supported actions (LVM only):

* create VG (Volume Group)
* add VG
* remove VG
* add PV (Physical Volume) to VG
* remove PV from VG
* create LV (Logical Volume)
* delete LV
* resize LV
* thin provisioning (overcommited disk space)

Not supported (use low-level tools instead):

* force LV to specific PV
* merge VGs
* split VG
* resize PV
* move content between PVs
* ...

### Encryption

* full disk encryptionn (LVM only: encrypted PVs)
* per partition (not root file system) with device mapper
* per partition also for the root filesystem (currently not supported,
AI: find out why it is not possible and fix it)
* using LUKS (Linux Unified Key Setup)
* limited support for legacy method: loop devices and crypto kernel


* create
* mount
* unmount
* /etc/crypttab and /etc/fstab handling

Not supported (use LUKS commands instead):

* LUKS multiple password slots handling
* change password
* set cipher (encryption method)

### Multipath

* network storage devices (typically fibre channel) where multiple
low-level kernel devices for the same storage device exist that have redundant
connections to the machine
* improved performance (parallel I/O over multiple channels)
* fail safe against some I/O channels being blocked
* special hardware that needs to be set up with special tools (outside
the scope of YaST storage)
* for most operations, it is important to use only the one device mapper
device for that storage device (representing the multipath device), not the
multiple low-level devices (representing individual paths to the device)

Supported actions:

* start
* stop

### Multiple Dimensions

* Btrfs (which has its own volume manager) on LVM
* Btrfs on LVM on RAID
* Btrfs on encrypted LVM on RAID
* ... (use your imagination)

## Remote Storage

This covers storage devices, which are not accessible out-of-the-box and some
operation is necessary in order to bring the disk to the system (and create a
/dev/.* device)

### SystemZ Configuration
(s/390 specific)

* DASDs (Direct Access Storage Devices) need to be activated for use
with Linux with the *dasd\_configure* command
* DASDs are created/managed by the s/390 host system
* Ihno wants to provide a script which lists all disks
* zFCP
* AI: Provide specifics for zFCP devices

### iSCSI
(Internet SCSI)

* networked storage using SCSI commands over TCP/IP
* using a separate set of commands and tools
* currently out of the scope of YaST storage, using a separate YaST
* might need to be integrated into YaST storage some day

### FCoE
(Fibre Channel Over Ethernet)

* networked storage using fibre channel over ethernet
* using a separate set of commands and tools
* currently out of the scope of YaST storage, using a separate YaST
* might need to be integrated into YaST storage some day

## File Systems

File Systems and operations to be supported by new libstorage

| Filesystem | Create | Root | Resize | Comment
| ---------- |:------:|:----:|:------:|
------------------------------------------------------- |
| ext2 | X | X | X |
| ext3 | X | X | X |
| ext4 | X | X | X |
| Btrfs | X | X | X | primary root parition, avoid for data
| XFS | X | X | X | preferred for data
| JFS | X | X | X | Drop
| ReiserFS | | X | X | Support existing partitions, not
allowed to create new |
| FAT/VFAT | X | X | X | for EFI system partition, else? AI for
PM to clarify |
| NTFS | | | X | AI for PM
| tmpFS | | | | ignore in libstorag (except mounting)
| HFS | | | | drop
| HFS+ | | | | drop
| NFS | | X | | installation on NFS as root is
supported |
| ZFS | | | | Not supported
| GFS2 | ? | | | AI storage people
| ceph | ? | | | AI storage people
| ocfs2 | ? | | | HA, AI lmb

Supported actions (for all filesystems, unless excluded in table above):

* create
* resize (for some types, see above)
* /etc/fstab -handling (managing individual mount points, mount options)
* mount
* unmount

A file system can be located in a disk partition, in whole disk, on an MD
array, on a logical volume

## Bootloader Constraints
***Caution: Voodoo inside***

* MS-DOS - Legacy BIOS
* BIOS loads the first sector of the disk (the MBR (Master Boot
Record)) and executes it
* only 512 bytes - sizeof( partition\_table ) = 446 bytes for this
first stage of the boot loader, i.e., the next stage (Grub2, Grub, LILO) has to
be chain-loaded
* the next stage has to be accessible by the BIOS
* limitations depending on BIOS:
* legacy: 1024 cylinders in CHS (Cylinder/Head/Sector) mode
(obsolete today)
* LBA (Logical Block Address):
* 32 bit LBA (obsolete today): 2 TB limit
* latest (7/2015): 48 bit LBA; 144 PB (144000 TB) limit
* there have always been limits, they always seemed way out of
reach, and we always reached them before we thought it possible.
* BIOS can only access first 8 disks
* Stage1 and Stage2 of grub2 should always be on the same disk
(means including the partition where stage2 will be insalled into and the
* BIOS can only support 8 harddisks, so bootloader needs to be on
the first 8 ones

* UEFI (Unified Extensible Firmware Interface):
* typically used with GPT disk label
* requires EFI system partition (with FAT file system)

* PPC (Power PC)
* can use MS-DOS or GPT disk label
* requires PReP partition in KVM or LPAR, on LPAR less than 8MB,
idealy size of grub2 stage1
* no special partition required for OPAL/PowerNV/Bare
* ***TO DO: Get more information from the PPC arch maintainers***

* s/390
* uses a z/IPL (Initial Boot Loader for z Series) initial boot loader
* uses a /boot/zipl partition (typically with ext2 file system)
* AI: Define requirements on /boot/zipl partition

### Disk Labels
(ways to create partition tables)

* Allow restricting installation to a single disk, even if this is not
the first one (use case with many disks)
* On new, empty disks, if there is no partition or if the disks will be
wiped (use full disk): prefer GPT disk label

* MS-DOS (PC classic/legacy)
* can be used on all supported architectures: i586, x86\_64, ppc,
* 4 primary partition entries in MBR (Master Boot Record)
* one partition can be an "extended partition": A container for
*logical partitions*
* each logical partition links to the next one
* the first logical partition is always /dev/sd..5 (no matter if
/dev/sd..4 etc. exist)
* need bootflags to mark active partition for BIOS

* GPT (GUID Partition Table; new PCs with UEFI boot support)
* New tool of choice: gptfdisk / sgdisk
* Backup copy might be present or not (not mandatory)
* partitions can have labels
* can be used on all supported architectures: i586, x86\_64, ppc,
* large number of partitions (128 or more)
* no longer limited by available space in MBR
* partitions keep their numbers when other partitions are removed or
* parted will use the smallest available number when creating a new
* bootflags are not required (maybe important for legacy BIOS,
ignored UEFI)
* **UEFI**
* EFI system partiton, vfat formated, might require specific
UUID, limited to 32GB (vfat)
* **legacy BIOS**
* protective MBR (should have active flag)
* hybrid MBR -> we do not want that any longer
* GPT Bios grub partition (needed for grub2)

* DASD (s/390 specific)
* Up to 4 partitions
* Separate zipl partition needed
* First sector is special, only writeable with appropriate tools
* Needs to write the blocklist onto zipl partition, kernel/initrd
must reside there (needs to be large enough to hold two copies of kernel and
* AI: Provide details about specific tools for first sector writing,
reasoning for two copies of kernel (why two?) for DASD

* Not supported or very limited support (which will be dropped in the
future) in YaST storage:
* Fixed number of 8 or 16 partition slots
* Fixed role (by convention) for some slots:
* slot a: root
* slot b: swap
* slot c: complete disk
* Sun
* Mac
* Amiga

### Dedicated Boot Partitions
(AI: combine with disk labels and bootloader constraints; need to be clarified
before merging)
***very architecture/system/bootloader specific***

* when is a boot partition needed?
* on what disk should it be?
* where on the disk should it be? (Boot sequence!)
* size constraints? (too small vs. too large; ppc for example loads the
*PReP* partition into the RAM completely, so it should be as small as possible)
* dependent on boot loader type
* dependent on encryption (BIOS cannot load files from encrypted
* dependent on presence of LVM

## Automated Storage Proposal

***Caution: Voodoo inside***

* a lot of code that tries to make a reasonable suggestion for many
different storage setups
* trying to be everybody's darling (well, at least most people's)
Supported scenarios:

* simple partition-based proposal
* with or without LVM
* with or without Btrfs
* with or without LVM full disk encryption
* with or without automated file system snapshots (if using Btrfs)
* with or without SWAP partition
* using all available space or leaving up unpartitioned disk space
* using whole disk or keeping (and resizing) already installed system
* RAID-based proposal?
AI: speak with different parties which requirements they have (Cloud, SES, SAP,
* -> bring all the use cases up
* -> mark those that make most sense
* -> discuss the implementation of them with UX expert

Not supported by the automated proposal:

* will work for HW RAID or BIOS RAID, which both can be seen as a
single disk by proposal
* -> Options to be explored, so far we have no experience with
proposal spread over multiple disks

##Misc. Topics to be taken care of

* Support Systems with >> 800 hard disks
* btrfs subvolume handling
* btrfs snapshots
* btrfs rollback: for this we need to install into a btrfs snapshot already
* Use udisks for storage detection "to do it the right way"? => Not a good
idea, see
* Samba?
* SSD?

< Previous Next >
List Navigation
This Thread
  • No further messages