Wandering Thoughts archives

Oracle, ZFS, and Linux (and Solaris)

A commentator on my entry on GPL'ing ZFS raised a good issue:

But don't you think it would have been reasonable for Oracle to relicense ZFS upon buying Sun, so they could use it for their own Linux distributions and wouldn't have to reinvent the wheel and support BTRFS development?

In my view, the core issue for Oracle with this is that GPL'ing ZFS gives it to all Linux distributions, not just Oracle's; it's an even greater gift to all Linux distributions if Oracle then also ports ZFS into the Linux kernel. Such a move also does some amount of damage to the remaining attractiveness of (Oracle) Solaris. All of this still might make sense for Oracle to do, but it depends a lot on how Oracle sees both their Linux distribution and Solaris.

I think that Oracle doing this makes the most sense if Oracle sees Solaris as a sideline and expect to be the leading commercial Linux distribution; then any harm to Solaris is unimportant and while ZFS benefits all Linux distributions, Oracle gets the largest benefit. Reversing this gives us the worst case for Oracle, where Solaris is supposed to be as competitive as possible and Oracle Linux isn't a serious player.

I basically have no idea where Oracle is going with Solaris. However, I don't think that Oracle expects Solaris to be big; the time of commercial Unixes has basically passed and Oracle's moves with Solaris don't seem calculated to increase its popularity. There might still be significant money in large enterprises and legacy customers, but even if so I doubt many of them would jump ship to Linux just because it now has ZFS.

I'm sure that Oracle would like Oracle Linux to be the biggest commercial Linux distribution, but at the same time I'm not convinced that that's likely or that Oracle expects to get there any time soon. Also, GPL'ing the ZFS code isn't immediately helpful to Oracle Linux (or any other Linux distribution); it only becomes helpful once ZFS gets integrated into the kernel, and that would probably take years (and have to be funded).

Given all of this, I've wound up thinking that GPL'ing ZFS lacks a clear benefit for Oracle (well, one beyond good PR). Especially, it doesn't solve any problems for Oracle; both ZFS and BTRFS would need a good amount of Oracle's engineering effort before they became ready for production. While it's true that someone else might integrate ZFS into the Linux kernel, saving Oracle the effort, it's equally possible that someone else would do a bunch of work on BTRFS.

(Given that helping create a new filesystem is more interesting than the drudgery of porting a large complex chunk of Solaris kernel code to Linux, it's probably easier for Oracle to recruit good kernel programmers to work on BTRFS than on a hypothetical ZFS Linux port.)

Can Oracle make ZFS much more attractive?

In my ruminations on the future of ZFS I called the Solaris version of ZFS dead due to the combination of the departure of the original ZFS developers, the (now) closed source nature of Solaris, and Oracle making Solaris unattractive to people. A commentator took me to task, noting:

We may not like Oracle, but they can afford developers and I do know they are working on ZFS. There are already new features in Solaris 11's ZFS that are not in Illumos, like: encryption, new ARC with in-memory "dedup", 1MB recordsize, better integration with fmtopo, etc.

The commentator is right; I was overly aggressive here. Oracle's version of ZFS is continuing to evolve (at least for now) and so at that level I'm clearly wrong. The bigger question is whether this matters. The core reason I had for it not mattering is Oracle's pricing and general attitude on Solaris making it quite unattractive, but I don't know for sure how much the price et al matters to people. (I know how much it matters to us, but I shouldn't generalize too much.)

However I think we can make some estimates from two things. First, Solaris in Sun's hands at Sun's (low) prices was not exactly setting the world on fire with its popularity. Second, it seems highly unlikely that Solaris got significantly more attractive when Oracle bought Sun (and certainly it's still not setting the world on fire). Given the significant price jump I think it's reasonable to believe that Solaris usage and attractiveness has dropped significantly.

(The potential fly in the ointment is if most or almost all Solaris customers were price-insensitive. I don't think that this was the case but I don't have any particular evidence for my belief. We're certainly price sensitive, but that's just an anecdote.)

This gets me to an interesting question: could Oracle make its version of ZFS matter by adding some features that are really attractive, attractive enough that people run Oracle Solaris in order to get them?

(If they can't and use of Oracle Solaris is basically static and small, then Oracle's ZFS is basically an unimportant backwater.)

My tentative answer is that I don't think so; I can't see how Oracle can make ZFS (in Solaris) significantly more attractive than it already is. ZFS just doesn't seem to be missing anything big, and what's being added is just things that are nice to have but not essential. ZFS today is not incomplete software, which should not be surprising given that Sun put years of development and polish into it. I just can't think of any crucial features that ZFS is currently lacking that would entice new people to use it (much less pay a lot of money to do so).

(This is of course related to the lack of enticing new features for us. It's possible that I'm not imaginative enough or that I'm missing the importance of some features to people.)

As a side note, although I would dearly love to have the ability to remove a vdev from a pool I'm unable to convince myself that it's such a crucial missing feature that adding it would make ZFS significantly more attractive. The same is true in spades for a 'ZFS fsck' and repair tool, partly because most people assume that their filesystems are never going to break in the first place.