Where is the ZFS?

[Nicholas Savage]

Why ZFS? Because!

Huh.

For me, the disk integrity features of ZFS are a part of my interest in using it. The rest is raid/caching/flexibility/cost-savings related. I can get 2x consumer/nas sata drives for the price of nearline SAS or raid drives. It is stupid easy to scale up/down to adjust for client needs. Getting raid and storage redundancy that is hardware agnostic is beyond awesome .

I have already had to recover a big raidz pool when someone's sas host card fried. . .attached the drives to two SATA FIS switching port replicators; brought the pools up. Used that hardware until the replacement raid card came. Have fun doing that when your synology box implodes.

Bah. Anyway.

On another note, Apple needs an nicely integrated iSCSI Initiator.

 

[grahamperrin]

Cows, CoW, Btrfs, performance of ZFS and the OS

File systems dairy

… I use butter I'm sure the butter team could use Apples resources just like I'm sure the the ZFS crew could use them too. …

For any reader who is puzzled by the reference to dairy produce: this one is unrelated to cows. butter F S is a pronunciation of Btrfs – another type of file system that uses CoW.

For any reader who is puzzled by the uppercase C and W: moo

Performance of ZFS with and without RAID-Z, on OS X and elsewhere; and RAID

… ZFS … performance it is very very slow.

To clarify. The old array was a simple 8 disk raid 5.

A few points.

Unlike snarfquest, I never worked with a 74.x version of MacZFS so I can't comment on its performance.

https://diigo.com/01cjz0 "… RAID-10 is preferable to either RAID-5/6 or RAID-Z for high performance work such as video editing …". That's taken from my ZFS RAID collection of bookmarks, which probably includes links to stuff about comparisons with RAID-5, but see below.

At this point in time I expect neither ZEVO nor ZFS-OSX to be optimised for performance across the broadest possible range of use cases.

That said, since I began using the tunables within ZEVO – around two weeks ago – there's a remarkable increase in performance (a MacBookPro5,2 that's unusually heavily loaded with data).

I understand that the state of development of OS X is not universally pleasing to developers, but from my customer perspective:

• Apple has done a amazingly good job with Mavericks

– more than with any past major upgrade (Leopard to Snow Leopard, Lion to Mountain Lion and so on), the difference in performance between Mountain Lion and Mavericks is remarkable. YMMV, personally I'm delighted … occasional wow moments. I always argue that benchmarks alone can't do justice to the quality of an OS. Holistically, the performance of Mavericks is very pleasing and hopefully a sign of better things to come.

(Performance aside: I haven't used 10.9 enough to recommend it without hesitation. My hesitation is likely to cease after 10.9.1 is released.)

OK, what about performance of ZFS with (not in) Mavericks?

When I have something substantial to report, it'll probably appear in the rally area. In the meantime, https://twitter.com/grahamperrin/status/405121768307703808 includes a link to a presentation from the OpenZFS Developer Summit.

I assume that "data-driven, adaptive algorithms" – as a potential next step for OpenZFS generally (not for OS X in particular) – could mean an end to people like me wondering about the pros and cons of tunables. If Apple is to go with ZFS/OpenZFS, I'd ask the company to accelerate this step.

File systems diary

Dates? No! Dont ask! For me it's Mac Pro first then I might speculate about what's next on the road/rally to 10.10.

 

[Freyqq]

HFS+ and FAT32 are the only file systems that I've ever had to deal with corruption and lost data. NTFS has always been rock solid. I look forward to a revised filesystem from Apple in the future.

 

[grahamperrin]

Back for a moment to page two, the other Microsoft example. I'm reminded that ReFS alone (without Storage Spaces) can not automatically correct corruption.

… NTFS has always been rock solid.

With the same respect that's in post 35 on page two, I'd ask any user of NTFS how she or he verifies the integrity of all files, all data, all metadata within the file system.

With hundreds of computers running Windows in my work area (thankfully not all nearby) I sometimes get a sense, without thorough investigation, of when NTFS is less than solid. CHKDSK might not reveal anything, but when a machine's playing up in a certain way I can guess that the disk will be on its way out before long. A few months ago, the guesses soon proved true for four of five heavily used machines in a student lounge. (I could/should have been more proactive about thoroughly checking the disks with something like HDAT2 (on UBCD) – long before the failures – but I knew that we'd get new machines before long so I was quite blasé about it.)

I look forward to a revised filesystem from Apple in the future.

+1

 

[MyMac1976]

HFS+ and FAT32 are the only file systems that I've ever had to deal with corruption and lost data. NTFS has always been rock solid. I look forward to a revised filesystem from Apple in the future.

Stability is not what people who look to ZFS and Btrfs are looking for. I've not had an unstable file system since the early 90's it's more redundancy.

Here is what btrfs says about itself in it's Wiki

Btrfs is a new copy on write (CoW) filesystem for Linux aimed at implementing advanced features while focusing on fault tolerance, repair and easy administration. Jointly developed at Oracle, Red Hat, Fujitsu, Intel, SUSE, STRATO and many others, Btrfs is licensed under the GPL and open for contribution from anyone.

The neat part about both Btrfs and ZFS is the ability to roll back to a known good state. I use Btrfs because I use Arch as my primary OS, Arch sits on the bleeding edge and if there is a problem I want to be back immediately. Btrfs was intended to be the ZFS of Linux, and it accomplishes much of what ZFS does but in a more clumsy way. I think it'll be as effective as ZFS once it's been around as long as ZFS.

HFS+ dates from the '80's it's time for Apple to move on ZFS wouold probably be the easiest but....

 

[rekhyt]

I move TBs of data around all day long and I've never had a corrupted file, ever. I can replicate the German method to corrupt a file by using dd, but that is really academic in my opinion. I understand ZFS is technically better, but the reality is that for the average user, and even pro users, there's really nothing wrong with HFS+. Maybe...MAYBE if you have a transactional database running on a server this would be a little more of a concern. But data integrity has been wildly overblown as a reason to move from HFS+ to ZFS.

I'm a novice user so I don't really know much about the theory behind filesystems like this (I've been trying to follow the thread avidly though) and data corruption is one of my biggest fears (although it hasn't happened yet).

I was wondering whether you could tell me how I could determine the integrity of my data on my harddrives.

Thanks.

 

[grahamperrin]

Verifying integrity with HFS Plus, with ZEVO

… how I could determine the integrity of my data on my harddrives. …

For HFS Plus: diglloydTools IntegrityChecker can be used to detect future corruption, but not to detect corruption that may have occurred before use of the product.

Related, in Ask Different: How should a user of OS X verify whether backed up data comprises everything required for lossless restoration?

If/when there appears a version of ZEVO for Mavericks, then you'll be able to give all or part of a disk to ZFS – instead of HFS Plus.

In ZFS terminology: a pool can be scrubbed; this verifies the integrity of all data on all disks within a pool. You can start and stop a scrub with the ZEVO pane of System Preferences:

 

[grahamperrin]

The disk you inserted was not readable by this computer

What are you referring to? I'd love to know. …

One of the things …

With a clean installation of Mavericks – without ZEVO

If you attach a disk that was used with ZEVO Community Edition 1.1.1, you should not see a prompt to initialise the disk.

The yellow alert (screenshot attached):

• does still exist in Mavericks
• does not apply to ZEVO data disks (I can't recall how the OS responds to attachment of cache (L2ARC) disks from ZEVO).

Simply: it's not appropriate to suggest immediate destruction of what may be valuable data.

On its own, this enhancement to the operating system is not cause for excitement in this topic. When I last tested, a few months ago, I found that the enhancement was not specific to ZFS.

 

[Glassed Silver]

Holy schmokes, seems you folks have been quite active since the last time I was in this thread.
I'll catch up later, but this straight away:

One of the things …

With a clean installation of Mavericks – without ZEVO

If you attach a disk that was used with ZEVO Community Edition 1.1.1, you should not see a prompt to initialise the disk.

The yellow alert (screenshot attached):

• does still exist in Mavericks
• does not apply to ZEVO data disks (I can't recall how the OS responds to attachment of cache (L2ARC) disks from ZEVO).

Simply: it's not appropriate to suggest immediate destruction of what may be valuable data.

On its own, this enhancement to the operating system is not cause for excitement in this topic. When I last tested, a few months ago, I found that the enhancement was not specific to ZFS.

Uhm, it's been a long day today and my head is filled with my Italian lessons, so I assume I'm not quite getting where you're getting at here.
What's the enhancement precisely?

The box that asks whether one wants to initialize or cancel (/continue anyway)?

For HFS Plus: diglloydTools IntegrityChecker can be used to detect future corruption, but not to detect corruption that may have occurred before use of the product.

[...]

That's awesome, I've been looking for something like this.
Are there free or less alternatives, too? Maybe open source even?
40 bucks for a good software sure isn't too much, but this is merely a 1-year license and with OS X being on a yearly upgrade path, I'll wildly assume that yearly updates to such a software is advisable.
I know it gets dramatically cheaper for 2 years and up, but a) 40 bucks is a lot of moolah for me right now and b) whatever the sum, 40 or 55 or whatever, before throwing such amounts of cash at a developer I want to be convinced by their product. It's a commitment past the worries of licensing fees.

Glassed Silver:mac

 

[grahamperrin]

What's not seen, reading between the lines; where IS the ZFS?

… I'm not quite getting where you're getting at here.
What's the enhancement precisely?

The box that asks whether one wants to initialize or cancel (/continue anyway)? …

I do find it difficult to describe. The point of interest is what's not seen.

An example. If you use ZEVO with Mountain Lion then upgrade to Mavericks: the disk is readable, the ZFS file system is not.

The disk is basically readable by the operating system – amongst the slices, in a partitioning scheme that's Mavericks-friendly, the OS recognises the presence of a ZFS slice. It's reasonable to assume that the slice will contain data, so logically:

• the box (the yellow alert with an option to initialise) does not appear.

The file system is not readable, which brings us full circle to the primary question in this topic:

• Where is the ZFS?

… quite active …

My own activity aside: I hear rumours about rumours but these are not yet good enough for MacRumors.

… 40 bucks for a good software sure isn't too much, but this is merely a 1-year license …

Indeed it's not costly. But if you contemplate an add-on to HFS Plus because ZEVO Community Edition 1.1.1 is incompatible with Mavericks and/or because ZFS-OSX is not yet stable: I would suggest waiting a while.

 

[Glassed Silver]

I do find it difficult to describe. The point of interest is what's not seen.

An example. If you use ZEVO with Mountain Lion then upgrade to Mavericks: the disk is readable, the ZFS file system is not.

The disk is basically readable by the operating system – amongst the slices, in a partitioning scheme that's Mavericks-friendly, the OS recognises the presence of a ZFS slice. It's reasonable to assume that the slice will contain data, so logically:

• the box (the yellow alert with an option to initialise) does not appear.

The file system is not readable, which brings us full circle to the primary question in this topic:

• Where is the ZFS?

My own activity aside: I hear rumours about rumours but these are not yet good enough for MacRumors.



Indeed it's not costly. But if you contemplate an add-on to HFS Plus because ZEVO Community Edition 1.1.1 is incompatible with Mavericks and/or because ZFS-OSX is not yet stable: I would suggest waiting a while.

I'll do this in 1,2,3

No. 1
I see, well that's interesting indeed. I see what you mean now and maybe it's something.

No. 2
Good enough for me, I love me a good rumor [about rumors] when it comes to ZFS and the Mac. Bring it on, bro, I read your warning note now, time for the rumors! If you don't want to be that guy who spreads rumors on rumors that are vague and even too shakey for MR, then drop me a PM.

No. 3
Hmm, maybe, but meanwhile I'm sitting here continuing to shiver away at the thought of losing valuable data...

Glassed Silver:mac

 

[grahamperrin]

Sorry, no private messaging. When it's good for public consumption, it'll be duly publicised.

 

[grahamperrin]

Oracle to acquire GreenBytes

… I hear rumours about rumours …

Not what I expected but still, of interest:

… Oracle announced it has agreed to acquire GreenBytes, a provider of ZFS technology … GreenBytes' technology complements Oracle's storage solutions and is expected to enhance Oracle's ZFS Storage Appliances. …

Whilst I don't expect development of ZEVO to resume under Oracle, I do wonder what will happen to the code as a result of the acquisition.

 

[maflynn]

But didn't Oracle already have ZFS because it was originally developed by Sun whom they bought?

 

[grahamperrin]

De-duplication, OpenZFS, ZFS, ZEVO and so on

Oracle's statement this month draws attention to de-duplication, replication, and virtualisation – as areas in which GreenBytes has expertise.

Focusing for a moment on one of those three areas, rewind five years or so … Sun and GreenBytes disagreed over intellectual property relating to de-duplication in ZFS. That issue was settled in 2010.

Fast forward to 2013, a patent granted to GreenBytes:

• Method of minimizing the amount of network bandwidth needed to copy data between data deduplication storage systems

I guess that Oracle's primary motive for the current acquisition is potential improvement to de-duplication in, or for, Oracle ZFS … until the transaction is closed, I don't expect true details to be known.

Relevance to Mac users

Of the minority who are interested in OpenZFS and/or ZFS, few will have a reasonable use case for de-duplication. Still – because of GreenBytes expertise in that area – in the past, I sometimes wondered quietly whether an unreleased build of ZEVO included improvements around de-duplication. Whilst I'm not familiar with the technical details of what's patented by GreenBytes, something that I saw (a video?) made me doubt that possibility.

Features that are more relevant to Mac users: I could go fishing for answers, but I expect the cloaks of secrecy to be as great now as they ever were. At least until after WWDC 2014.

(If there'll be an opportunity to resume development of ZEVO, I doubt that any plans will be expressed until after Apple announces whatever it has up its sleeve for the thirtieth anniversary year of the Mac.)

As developer Don Brady says, it's sad to see ZEVO languish …

Postscript

Oracle buys desktop software virtualiser GreenBytes • The Register

> Gets better-than ZFS dedupe

> … could mean the ZFS appliance getting GreenBytes’ deduplication engine. …

 

[maflynn]

I think this continues to mean the odds are getting higher in not seeing anything for OSX with ZFS. The market is small enough for ZFS on the Mac and with Oracle buying the company the odds are imo they'll shut down anything that isn't core to Oracle's goals.

 

[grahamperrin]

ZFS use of memory on Mac OS X

I stumbled across an interesting LinkedIn profile …

Wenguang Wang | LinkedIn

"… Architect/Main Contributor of the ZFS buffer cache re-write on Mac OS X …"​

… it notes that ZFS buffer caching was dropped …

Not quite; the word was 're-write'.

OpenZFS on OS X

The topic below was posted, a few minutes ago, to https://openzfsonosx.org/forum/viewforum.php?f=24 … it disappeared during edition, I'll try to arrange a re-post. there's now more information in the wiki (thanks!).

----

Mavericks and greater: memory pressure sensor, ARC, reaping

Attempting to familiarise myself. Is the collection below a reasonable starting point? Thanks …

----

From https://openzfsonosx.org/w/index.php?title=Changelog&direction=next&oldid=731 (2014-10-17):

…(snip)…​

mach_vm_pressure_monitor

http://www.opensource.apple.com/source/xnu/xnu-2782.1.97/osfmk/vm/vm_pageout.h

No pressure, Mon! Handling low memory conditions in iOS and Mavericks (2013-11-03)

Also of interest

Consider re-adding illumos page out logic in arc_memory_throttle · Issue #210 · openzfsonosx/zfs · GitHub​

 

[SlCKB0Y]

Wenguang Wang | LinkedIn

"… Architect/Main Contributor of the ZFS buffer cache re-write on Mac OS X …"​

Not quite; the word was 're-write'.

OpenZFS on OS X

The topic below was posted, a few minutes ago, to https://openzfsonosx.org/forum/viewforum.php?f=24 … it disappeared during edition, I'll try to arrange a re-post.

----

Mavericks and greater: memory pressure sensor, ARC, reaping

Attempting to familiarise myself. Is the collection below a reasonable starting point? Thanks …

----

From https://openzfsonosx.org/w/index.php?title=Changelog&direction=next&oldid=731 (2014-10-17):


– and in the current edition of the change log, that line is without reference to SPL.

I see 'memory pressure' in comments at/around https://github.com/openzfsonosx/spl...948e6f014d2263c3379/module/spl/spl-kmem.c#L53 and https://github.com/openzfsonosx/spl...948e6f014d2263c3379/module/spl/spl-kmem.c#L75

The closing comment (2014-11-21) from Wired memory grows from less than 3GB to more than 10GB over several days - is it ZFS? · Issue #213 · openzfsonosx/zfs:



From https://github.com/openzfsonosx/zfs/issues/270#issuecomment-71737092 (2015-01-27):



Open issues

Amongst search results, this one caught my eye:

ARC release bug triggered · Issue #260 · openzfsonosx/zfs · GitHub

– the reporter uses OS X 10.8.5, most recently with OpenZFS on OS X 1.3.1-RC5. Mountain Lion can't benefit from the modern memory pressure sensor …

… is it too early to tell whether #260 might also occur with Mavericks or Yosemite? I'm curious.

mach_vm_pressure_monitor

http://www.opensource.apple.com/source/xnu/xnu-2782.1.97/osfmk/vm/vm_pageout.h

No pressure, Mon! Handling low memory conditions in iOS and Mavericks (2013-11-03)

Also of interest

Consider re-adding illumos page out logic in arc_memory_throttle · Issue #210 · openzfsonosx/zfs · GitHub

Why are all your posts almost unreadable and give the reader the sense of walking into a conversation late where there is only one participant (you)?

 

[grahamperrin]

Performance

Apple's continuing preference for HFS Plus

From another recent topic:

What, would you say, are Apple's reasons? …

Simple. Inertia, lack of burning need for anything better, and lack of any really obvious and obviously better alternative -- or at least, lack of an internal evangelist for an alternative.

I'm not trying to defend HFS+ except that it's been mostly good enough over the years.

As for alternatives: ZFS, maybe. …

From earlier in this topic:

… for the average user, and even pro users, there's really nothing wrong with HFS+. …

With hard disk drives: I strongly suspect that booting from (and running with) HFS Plus is detrimental to performance – compared to a more suitable file system.

However there's probably nothing known, publicly, about how OS X might perform with ZFS alone (without HFS Plus). I haven't seen OS X boot from ZFS, but it has been done.

Simply: users know no better. In that sense … yes, HFS Plus might be described as mostly 'good enough'.

Performance, with a ZFS home directory and with OS X booted from a different device

From https://openzfsonosx.org/forum/viewtopic.php?f=11&t=2171&p=4883#p4883 :

"… As far as I recall the Mac performed much better when the operating system was booted from a much slower hard disk drive (probably on USB 2.0). …"​

– my sense of things was that not using the HFS Plus slice of the internal SSHD allowed the encrypted Core Storage slice to 'shine' with ZEVO.

 

[grahamperrin]

OpenZFS on OS X use of memory

Noted with thanks, recently added to https://openzfsonosx.org/wiki/Performance:

"… 1.3.1 will by default allow the ARC to occupy most of your computers physical memory. In parallel with that it has a mechanism that detects when the computer is experiencing memory pressure and will release memory back to the OS for other purposes. This mechanism has been extensively tested, but may not be suitable for everyone.

It is possible to temporarily limit the size of the ARC …"​

 

[grahamperrin]

… Maybe open source even? …

A few days ago with OS X 10.9.5 I began testing recently released version 1.4.5 of OpenZFS on OS X. Preparing for possible production use of that open source software …

The situation has changed with the advent of stuff like OpenIndiana. Heck, FreeBSD itself uses ZFS now, as of 10.0-beta1!

Yeah. PC-BSD (based on FreeBSD) makes very smart use of OpenZFS.

Hey graham; I lurk over in the ZEVO forums from time to time. …

Whilst the GreenBytes forum for the product is long gone, I continue to use ZEVO. All things considered (see below), for me it's the best solution and there's Don Brady to thank.

From a topic that was originally about the RAID capability within OS X:

… Can't recommend OpenZFS enough at this point. If you need a local … most resilient …

When it worked, performance was good. However: for me, repeatedly, OS X 10.9.5 stopped responding with OpenZFS on OS X 1.4.5. In itself, the bug did not alarm me; I had five years' experience with AppleSeed so I was well prepared for feedback to, conversation with, a different group.

… keeping quiet about progress doesn't exactly inspire confidence …

Keeping quiet can be OK. Keeping quiet is sometimes a necessity (confidentiality and so on).

Apple's Seed Team was impeccable, as were most other individuals.

For me, the OpenZFS on OS X community is let down by peculiarities such as disappearing content. No explanations; I simply stumbled, day by day, across a series of disappearances. I publicly queried the losses, there was an invitation for me to discuss things in private but honestly, I can't have confidence in an environment where there's a lack of courtesy:

• it's not good manners to take a silent, throwaway attitude to other people's work.

To the original question, in 2013 –

Where is the ZFS? …

– I wish well to users and developers of OpenZFS on OS X but for me, that's not where it's at; can't be where it's at.

Keeping quiet is one thing. Throwaway attitude is another …