"Project Springfield" Is Red Hat's Effort To Improve Linux File-Systems / Storage

[S.Pam]

Btrfs has some performance issues with NVMe drives when compared to EXT4 or even NTFS under windows. Until those are sorted out I'm unfortunately ignoring it.
I think it does ok for standard SSD drives however.

Can you backup that claim?

[starshipeleven]

I could hunt them down for you, but I will just do the tests myself.

When you do please send a message or something so I get notified. I've yet to see any article about NVMe issues with btrfs.

It would be nice if BTRFS was speedy for NVMe because it has a free windows kernel driver, but I doubt much has changed.

what does this mean? does Sata or any other storage technology not have a "free windows kernel driver"?
Why having a "free windows kernel driver" for the block device is worth mentioning?
The issue has always been that non-Microsoft filesystem drivers aren't a wddm-approved and signed so you have to jump throught some hoops on your own to load them.

[elatllat]

... I really don’t see the point of Logical Volume Management on a workstation....

There are disadvantages to ZFS and brtfs (license,stability,speed when full,smr speed, database friendly, etc).
LVM and friends have none of those issues and do have many (all?) the advantages of those two advanced local file systems;
like adding space to a volume without downtime, snapshots before risky operations or backups, lvmcache, shrinking volumes to mess with other file systems like btrfs,ceph,etc.

[S.Pam]

LVM only supports shrinking if underlying fs supports it. Does it support conversion between raid profiles? Does it support uneven sized disks in raid modes? What about detecting data corruption?

[cynic]

The Fedora team should refocus on software correctness. I find their disk management tools to be a big disappointment every-time I go to use them. That includes install time ( buggy as hell), the graphical tools offered and the command line tools. By the way some of this stuff has been crap for several releases.

so, have you let them know about all those issues you've found? Have you openened bug reports about them?

as for workstation support I often wonder why they default to LVM. It is a workstation , I really don’t see the point of Logical Volume Management on a workstation.

well, it is good:

* if you need to add storage (every user sooner or later asks for it);
* for the ability to snapshot and allow consistent backups (workstation user will thank you later);
* for the ability to easily add SSD caching to the volume (workstation users will love you for that);

these are things that are useful in general and not only on servers.

[starshipeleven]

There are disadvantages to ZFS and brtfs (license,stability,speed when full,smr speed, database friendly, etc).
LVM and friends have none of those issues and do have many (all?) the advantages of those two advanced local file systems;
like adding space to a volume without downtime, snapshots before risky operations or backups, lvmcache, shrinking volumes to mess with other file systems like btrfs,ceph,etc.

LVM has only SOME advantages of btrfs/ZFS, it's just block layer after all, it does no RAID nor checksumming nor deduplication, and so on.
It's noticeably faster than btrfs, even with nocow, but only slightly better than a ZFS array without SSD caching.

Also SMR drive speed isn't better with LVM (if we are talking of drive-managed drives anyway).

[starshipeleven]

LVM only supports shrinking if underlying fs supports it.

the fs is in the LVM partition/volume, there is no "underlying" fs in LVM.

Does it support conversion between raid profiles?

LVM has no concept of RAID. You use it on top of Mdadm RAID.

Does it support uneven sized disks in raid modes?

no (mdadm does not, I mean technically yes you can but you are limited by the devices with smaller size, bigger ones can only use as much space as the smaller drive in the array)

What about detecting data corruption?

Again, no concept of corruption detection and fixing. You must use dm-integrity device manager target under it.
To make a btrfs-looklike you basically need a mdadm RAID with on top a dm-integrity volume with on top a LVM device.

If you want to compete with ZFS (aka have a decent RAM/SSD caching) you need to also add a bcache volume between dm-integrity and LVM device.

If all this clown car of stuff sounds a huge pain in the butt to set up and manage, yes, it really is. That's why RedHat in their infinite wisdom have started the Stratis project, to provide a single tool that orchestrates this mass of cats into a single coherent system that does not require a Veteran Unix Admin to set up and manage.

[ayumu]

Maybe they'll realize NILFS2 exists, is in the Linux kernel, has data and metadata checksums, does snapshots and, most important, is designed for Low Latency.

But that's overly optimistic, chances are they'll just continue to go full retard with that BTRFS trash.

[polarathene]

There are disadvantages to ZFS and brtfs (license,stability,speed when full,smr speed, database friendly, etc).
LVM and friends have none of those issues and do have many (all?) the advantages of those two advanced local file systems;
like adding space to a volume without downtime, snapshots before risky operations or backups, lvmcache, shrinking volumes to mess with other file systems like btrfs,ceph,etc.

LVM snapshots aren't as good as the BTRFS kind.

Stability? BTRFS is pretty stable these days, are you biasing this based on issues from years back or features that are marked as unstable(RAID 5/6)?

Speed when full? What issues does BTRFS have regarding that?

SMR speed...? BTRFS is one of the few filesystems able to perform well with SMR, you'd be using dm-zoned otherwise for filesystems that aren't compatible(with host managed SMR zoned storage): https://zonedstorage.io/linux/fs/

What license issue does BTRFS have?

It's database friendly afaik, just disable CoW for where you're using a database, can still leverage snapshots, alternatively just use another filesystem for large data with heavy writes if necessary, you don't have to assign a single filesystem to all your storage space.

[elatllat]

...concept of RAID...

https://man7.org/linux/man-pages/man7/lvmraid.7.html

...To make a btrfs-looklike you basically need...

...LVM and friends...

...also add a bcache...

https://man7.org/linux/man-pages/man7/lvmcache.7.html