4 Feb
2021
4 Feb
'21
11:46 p.m.
+1 to everything Dan said, including the "who cares if it works?".
I was definitely coming at it from the kernel perspective, I joined Sun
just after SunOS 4.0 came out; that was the release that made mmap/page
cache be a coherent thing, the buffer cache was almost completely gone,
I think it was still used for inodes and directories but no user data
was in there.
The amount of work that went into that was huge but the resulting kernel
was actually smaller (I think) and much easier to understand and maintain
(I know). And the kernel team was quite proud of that work, with good
reason.
Which is why it blows my mind that the same organization that saw the
value of have one way to do things, did a 180 and said the page cache
isn't that important. When I talked to the ZFS guys about it, their
plan was that ZFS was going to be the only local file system so it sort
of didn't matter. But that's nonsense, you are going to have tmpfs,
nfs, vfat, ntfs, whatever apple uses, etc. So you are going to have a
page cache for all of those, which means the page cache is never going
away, which means that mmap() wants pages (remember, you can mmap stuff
that isn't in memory, you need the hardware to fault for you so you can
go get it). The ZFS guys said making that work is too hard, we did it
in BitKeeper, it works fine.
So the combination of going back to something anyone with OS smarts
knows is a bad idea plus the fact that I know from personal experience
that they could have done in the page cache and still have compression
and XOR, yeah, I lost a lot of respect for those guys.
ZFS is very useful, it could have been that and played nice with the
page cache.
On Thu, Feb 04, 2021 at 11:32:31AM -0500, Dan Cross wrote:
On Thu, Feb 4, 2021 at 10:47 AM Will Senn
<will.senn(a)gmail.com> wrote:
--
---
Larry McVoy lm at mcvoy.com http://www.mcvoy.com/lm
[snip]
In response to the negative vibes around ZFS. [snip]
I think the discordance is around the semantics ZFS's implementation
implies. Larry's point about mmap() vs a buffer cache is entirely valid; it
took lots of people heroic amounts of work worthy of Greek sagas to bridge
the difference between the original buffer and VM page caches, but ZFS
says, "meh. too much work; not worth it." The practical implication of that
is that memory mapped IO (via `mmap`) is no longer coherent with file IO
(via `open`/`close`/`read`/`write`) without lots of work that both degrades
performance and add complexity.
The question that a lot of folks who use ZFS regularly ask is, "does that
matter?" And perhaps it doesn't: if I've got a file server sitting there
serving NFS, do I care what it's kernel is doing? As long as it's
saturating the network and disks, and it's reliable...not really.
(Incidentally, that was kind of the philosophy behind the original plan9
file server kernel...as I heard the story, the rate of change of the plan9
kernel proper was too high, so Ken split off the file server portion into
its own, special-purpose kernel, and it stayed like that for ~20 years).
Similarly, if I'm on the local machine and the required coherence code is
there and largely works, then again, perhaps as a consumer of the
filesystem, I just don't care. After all, one can still get work done, and
ZFS has a bunch of other features that make it very attractive, right? In
particular, it's very good at NOT losing my data, kernel purity be damned.
On the other hand, if we're discussing OS design and implementation,
(re)splitting the VM and buffer caches is a poor decision. One might well
ask, "why?" and the answer may be, "because it adds significant
complexity
to the kernel." This to me seems like the crux of the disagreement.
Satisfied users of ZFS might legitimately ask, "who cares?" and one might
respond, "kernel maintainers." If the kernel is mostly transparent as far
as a particular use case goes, though, then I can see why one would bulk at
the suggestion that this matters. If one is concerned with the design and
implementation of kernels, I could see why one would care very much.
Like many things, it's a matter of perspective.
- Dan C.