Did Wilson & Ahrens just say that ZFS no longer holds an edge as to data integrity?

[guermantes]

(This is not really a help-me question, it's more an invitation to discussion and enlightenment.)

So, when I have too much time on my hands (or if I am procrastinating) I sometimes challenge myself with things that fly slightly over my head, in order to see if I can grow a taller neck... And today I decided to watch a talk by George Wilson and Matt Ahrens on ZFS from 2018. And a big question mark appeared early on, when Wilson came to point 3 in the list of things ZFS was intended to remedy: data-integrity and self-healing. I peaked my ears as this was a main reason why I ended up in the world of FreeNAS.
I am not going to quote Wilson, but to me it sounded like self-healing was not such a big selling-point anymore (i.e., only a selling point historically). I knew btrfs did something similar but I still thought ZFS was "king" is this regard. Did I misunderstand anything or have other filesystems caught up? Or is he talking about manually running tools of various kinds, or something like that?

Link to the relevant position in the talk:

 

[HoneyBadger]

Basically, the rest of the storage world finally realized that having redundant copies of bad data is silly, and caught up to ZFS. ;)

The pace of technology and the available CPU/memory bandwidth made it possible to do this while still providing the level of performance desired, so it's been widely adopted.

ZFS might no longer be the only solution doing this end-to-end protection, but in my opinion it's still the best.

 

[jgreco]

There's lots of commercial systems that do various types of data protection, and there's been a move away from monolithic storage arrays to hyperconverged solutions that may or may not have data protection.

ZFS was ambitious at the time it was designed.

Btrfs is probably the only major free software product that anybody might consider a potential competitor to ZFS, but it has suffered for years as a semi-stable trainwreck and gained a bad reputation in its own community as a result.

The real challenge to ZFS is that it doesn't have a well-designed strategy to deal with things like storage tiering. It is a GREAT design for HDD based storage. It is less of a great design for modern mixed requirement storage.

 

[Mlovelace]

The real challenge to ZFS is that it doesn't have a well-designed strategy to deal with things like storage tiering.

Nimble's CASL architecture does this exceedingly well; but CASL also looks like what you'd get if you gave ZFS, and 10's of millions of startup dollars to a group of storage engineers. Which could very well be the case. Either way their SANs are some of the best I've used, but I think HCI is going to push SANs out.

Edit: Meant to say I've been doing some testing with VMware vSAN leveraging iWARP and HCI is a very compelling direction.

 

[jgreco]

Either way their SANs are some of the best I've used, but I think HCI is going to push SANs out.

Well, if we're talking about for VM storage or other truly SAN-ny things, yes. I think ZFS is likely to be the long term winner for the more difficult art of managing massive storage pools though.

It seems to me like the typical bitrot ZFS is designed to catch has become somewhat less common in recent years. This may be due to better detection/correction mechanisms inside drives. I typically see a read error or a correct block. That wasn't always so.

I do use a lot of RAID1 datastores (LSI 9271CV-8i) on hypervisors. Almost everything I run has redundancy, which has largely meant that OS read failures due to bad sectors have not really been an operational issue. I find complexity can be problematic, and some of the HCI solutions seem to be excessively clever to the point of danger. As an admin, I want to know that I understand what's going on underneath the sheets, and a lot of the HCI stuff fails the sniff test. I can see HCI as a win at large scale, but most of the sites I run have a dozen hypervisors or less.

The part that really drives me crazy is the nonredundant storage strategies some people are using...

 

[Arwen]

There are now several self healing file systems. But each has limitations...

• BTRFS - Linux only; Not completely COW; Checksums stored with block, not block pointer; Checksum is currently weaker; Never stablized; RAID-5/6 is a mess
• Hammer - DragonFly BSD only; Still being written; While it's open source, it's basically few or 1 developer
• B-CacheFS - Linux only, (IIRC); Still being written; No where near feature complete compared to ZFS, or BTRFS in Mirror only mode.

In my limited and humble opinion, it would be better if some of those developers gave up on their limited FS efforts, and helped with OpenZFS project.