I suspect with CPU’s being way faster than necessary for most people (heck, my two year old CPU is still more than fast enough) and external storage being cheap and easy to add, a lot of external storage will end up being sedentary. In fact, I wonder what percentage of laptops end up staying in one place – I suspect it’s significant.

- bill

If Apple created a default multi-file-system configuration for the specific purpose of using different snapshotting policies that would indeed take the burden off the user. But doesn’t Leopard already have a somewhat similar mechanism (I forget the name, but it supposedly allows one to roll data back to fairly arbitrary points) in HFS+ – and does it suffer from inability to tune it such that it doesn’t snapshot things like temporary files?

While I use separate partitions for Windows and data myself, disk sizes these days are such that I just don’t have problems juggling space between them (though if I needed to Partition Magic or one of the free Linux/BSD utilities would make it trivial to do). So I still consider ZFS’s flexibility in this are to be more significant for corporate environments (where storage space must be juggled among multiple competing users and departments) than for the desktop.

- bill

I’m making no more assumptions about you than you making about me Anyway it’s not that important.

Multiple file systems – yep, you’re right – most desktop users don’t use them or use just two (quite a common thing on windows is to have c: for system and another d: for data). The probel is that creating those file systems couses too many problems for desktop users like: I do have a space in one file system while I run out of space in another… so one big file system is easier to cope with.
The think you didn’t get is that if Apple would create one zfs pool for system disk then several separate file systems (for temporary files, for intermediate downloads, for documents, etc.) then user wouldn’t even know about it as he could still see all available space no matter where he wants it. But then Apple can enforce different snapshoting policy for those different filesystems, so your documents or media files and system itself is snapshotted quite frequently while temporary files are not snapshotted at all. It would be totally behinde the scenes from the user perspective. Then if user wants to create a “directory/folder” with different snapshot interval what would happen in practice is that new zfs file system would be created with different snapshotting policy. It’s that simple.
If you think about the old way of doing things – you’re right, too much hassle. With zfs it’s just natural. So having multiple file systems is actually very useful on desktop and user wouldn’t even know that he is actually using multiple file systems.

Performance of snapshots – yep, most of solutions in a marted are using CoW but in a different way – so if you’re going to change some files the performance would be actually impacted as you’ve got to wait for CoW which is generally not the case with ZFS.

Guess how many of these machines came with more than a single hard drive standard? Guess how many typical users go to the trouble of ordering multi-drive systems, let alone adding an internal (i.e., fixed) drive later themselves? Guess how difficult it would be even to *find* a multi-drive system if you walked into a Best Buy or similar establishment (as most purchasers do) rather than configured one deliberately on line (as most on-line purchasers don’t)?

2. Your claim that computers are just about to become the focus of home entertainment might be a bit more credible if people hadn’t been trumpeting it for nearly a decade now. In fact, special-purpose devices have started hitting the market precisely because computers have *not* made good on this promise, and show no signs of doing so (for whatever reasons: some people just don’t want a computer in their living room, others had enough problems ‘programming’ their VCRs that the idea of an even more complex environment both in terms of wiring and of control appalls them, and in general a small, quiet box with a few simple buttons and plugs on it seems far more acceptable).

3. Pertinent to both of the preceding points is that laptop sales volumes are rapidly starting to dominate conventional desktop sales: just how prevalent do you expect multi-disk (especially beyond simple mirroring) laptop configurations to be, and just how popular do you expect laptops to be as the focus of home entertainment centers?

4. But it’s a lot less unrealistic to assume that *some* kind of storage-intense home entertainment usage will occur (just not much of it involving general-purpose desktops): what special advantages would ZFS bring to it? Closed ‘appliances’ would simply use matched drives if they wanted to support mirroring (leaving ZFS no value to add – people really aren’t going to want to have to ‘manage’ this storage, e.g., by mirroring material selectively) and use external USB drives for supplementary storage (no way for ZFS to add value there, either), especially for those people who might like to take their media files on the road once in a while.

5. Now, as a techie I can certainly see the value in having some kind of central home storage server that satisfies the needs of the home’s computers *and* its entertainment center, but I’m not sure that even that is likely to occur – even if wireless networking eliminates would otherwise be a wiring mess. Multi-use storage means that people have, at least to some degree, to understand how its multiple uses interact, and experience suggests that most people would be happy to pay something of a premium not to have to (i.e., to use separate appliances). If we get to the point where storage really does become a utility over the Internet and people don’t have to deal with it at all, then ‘home’ storage will become remote large-scale server storage and ZFS will start to become more interesting – but I don’t think that’s at all what’s been being discussed here.

Thinking about the evolution of audio equipment might provide some insight: my impression is that modular components peaked during the ’70s, after which most people save for audiophiles reverted to multiple stand-alone special-purpose boxes, even though they arguably provided inferior quality at a higher total cost. But I’m hardly well-acquainted with that market, so my impression could be completely wrong there.

Robin’s a big boy, Adrian: I expect (or at least hope) that he knows what he’s asking for by voicing his opinions as assertively as he does, which of course is why I’m similarly assertive when I disagree with him because – as I noted at his ZDNet blog – it sometimes takes liberal use of a heavy, blunt instrument to get his attention if he’s on a tear. The main faults I find with him are technical, since his depth in that area, while usually pretty respectable, sometimes just doesn’t match (or adequately support) his enthusiasm.

By contrast, your claims above seem to fall far more into the realm of uninformed speculation (you know what they say about opinions…) and I’ve responded in kind.

- bill

To say that that such things are just not important to average users gets average users completely wrong: even average users will tomorrow be making demands on their file systems that today look geeky. Robin understands that, Apple understands it, but somehow it is beyond the dirt-under-the-fingernail-tech guys here, who follow this issue around the net insisting that they, and only they, are right.

Not only is ZFS not being over-hyped here, in my view it is being undersold. It is needed on the desktop now, not in two years time. It’s failure to appear in Leopard is a significant disappointment.

You’re correct that the Google study neither addressed the incidence of ‘silent errors’ nor compared (S)ATA failure rates with FC/SCSI. The CMU study at the same FAST did the latter and found little difference; I don’t know of any studies that have explicitly targeted ‘silent errors’, but in the absence of evidence to the contrary (by all means present some if it’s not purely anecdotal) suspect that such errors likely follow the same pattern as other failure modes.

As for backup, you continue to confuse corporate environments (‘most shops’) with the subject under discussion here (home media servers). While the former have reason to perform full backups once in a while because their data *changes* and thus would require application of all incremental backups since the last full backup to reconstitute it, the latter do not (because data is *additive* and hence there’s no reason to do more than add it to the existing backup copy – which then becomes the current full-backup copy). Had you paid closer attention to the detail that I provided previously you would have understood that.

Furthermore (still with respect to backup) you’ve wandered away from the subject yet again: your early response was not about silent-error *file* corruption (which media files tolerate just fine for the typical case of single- or few-bit errors) but about the supposed hazard of catastrophic *metadata* (structural) corruption in the file system that would lose you the whole thing (Robin’s original doomsday scenario). If you’ll revisit my response with that in mind you will see that there’s no possibility of propagating such errors (in the unlikely event, I will remind you, that they occur before a meteor destroys your system) to your backup copy (and if every once in a while a bit in a backed-up media file is flipped, it will be profoundly irrelevant).

As for the personal value of end-to-end checksumming on your own desktop (I do hope that’s what you talking about), knock yourself out – just don’t expect the rest of the world (at least that portion which has any grasp of probability) to share your paranoia.

Since my observation that snapshot support was hardly unique to ZFS you seem to be trying to change the subject to using them with multiple file systems. If you create multiple file systems on your desktop so that you can snapshot them differently, that already puts you in a completely negligible minority of desktop users: when discussing the market relevance of that particular feature (which, I remind you, is the subject here) rather than just plain old whole-system snapshots, that minuscule percentage just doesn’t count. As for performance, I think that all the examples I presented perform copy-on-write of one form or another, rather than the antedeluvian mechanisms of the early ’90s where major metadata copying occurred when the snapshot was created, so performance is adequate, space use is reasonable, and at least for the open-source products I mentioned they’re free of charge as well. Writable snapshots for typical desktop use? You really do appear to be getting desperate here. Ditto for compression: with today’s processors compression (even good old Windows-style compression) can actually be a performance win, but disk space is so cheap that virtually no one bothers (just how much email are you talking about, anyway?).

That about covers it. I’m happy to leave it to anyone who’s sufficiently interested still to be reading this to assess which of us is being more objective (not that I care all that much: I tend to discuss things with people I disagree with more to try to elicit information that I may not be aware of, but in this case the effort is not proving very worthwhile).

- bill

-No, I think the fact that technically superior products – and let’s be real, not all technical “superiority” is something people will pay for – lose to inferior is NOT a good thing. But it is a reality.

Which is why it drives me nuts to see companies with good products assuming they’ll win because of their product. No, chances are good you will lose, unless you get good marketing.

Robin

I’m trying to not take this out of context, but do you honestly mean that? Great marketing for something with lacking technical merit is better for who exactly???

Google study wasn’t about silent data corruption on FC/SCSI disks IIRC it was about disk failure rates. If you want other users experience with silent data corruption on desktop servers – look at google, you’ll find them. My experience is quite good – I admit. Just one corruption in 2 years or so. Nevertheless no harm was done thanks to ZFS.

Backup – well most shops do actually full backups once in a while. I know TSM approach is to always do incrementals but that’s just one solution. And even if you backed up files soon after they were written the problem is they could already be corrupted… like in a case with failing SCSI card I encourted two times so far.

End-to-end checksumming for me is a must these days – and what you’ve got to pay for it in terms of CPU/disk IOPS is irrelevant in almost all cases in practice.

Snapshots on desktop – can you easily create sub filesystems in other solution so you can setup different snapshot policies for different “directories”? For example I don’t want a temp files or download directory to be snapshotted too at all, however I want my email to be snapshotted at least once a day (it is) the same goes for other data. With pooled storage approach it’s just so easy to achieve it and you don’t have to worry how you partition your disk or what size of your file system should be, etc.
Then what is a performance impact of using snapshots in other file systems? In ZFS it’s basically 0. Can you create a clone (writable snapshots)? What’s the disk space and performance impact? How easy is it to use? Do I have to pay for it?

Can you enable compression just for your emails or My Documents or whatever? I can and I do for my email – light lzjb so I don’t notice any CPU overhead and get 2:1 ratio.

I’m afraid your dislike for ZFS made you less than objective.
From my perspective I’ve been using VxVM/VxFS, LVM, UFS, SVM, ext2, ext3, xfs, …
And my experience (both desktop and server) with ZFS comparing to all the other solutions made me that enthusiastic about ZFS – it’s been making my life easier and solved lot of problems.

Apple has been clever picking up ZFS as its next generation file system. That some people don’t get it – well, it’s just the wayit is.

ZFS’s ability to detect otherwise undetected errors via end-to-end validation is actually a technical feat of some note, given how many other systems can’t and given how frequently high-end environments like EMC’s Symmetrix use similar mechanisms but only within their block-storage boxes, rather than end-to-end from host RAM to disk and back again. Since WAFL has been doing something very similar for many years it doesn’t qualify as great innovation, and since their underlying mechanism that facilitates the required in-parent checksums is essentially ’60s shadow-paging technology on steroids that isn’t tremendously innovative either (nor a very efficient use of disk bandwidth, given the number of pages that must be written back on every update: WAFL can do somewhat better here, since it can accumulate updates in its NVRAM and IIUC must update only block pointers rather than the full path to the root when it actually does write data back to disk), and since the incidence of additional errors detected is very low it’s not all that important save in very controlled and critical environments where they aren’t swamped by other kinds of errors – but it’s still worthy of respect.

Robert:

Please remember that the subject under discussion here involves desktop storage, which is rather unlikely to experience any of the firmware problems you encountered in Sun, IBM, and EMC storage because virtually no desktops use it. Desktop ATA/SATA host-controller hardware is dirt-simple and very standardized – desktops aren’t even likely to be using SCSI. In fact, your own desktop experience (though anecdotal) tends to support this – as do the only quantitative studies that I’m aware of.

I’ve always acknowledged that *in controlled server environments* (such as your Google example) ZFS’s added integrity can become significant – though I don’t necessarily agree that it’s notably more significant for ‘cheap’ SATA disks than for enterprise disks, and (again) the two recent studies discussed here – one by Google itself – tend to support this (contrary to your own anecdotal experience).

For a media server, “what’s in your backup” is presumably the media files soon after they were added: in such an environment there’s no reason to back them up again and again, because they never change: you just add the new ones to the existing backup disk. Furthermore, by using such an incremental, file-based backup there’s no possibility that a disastrous metadata error on the original system can propagate to the backup copy, so the worst that can happen in the unlikely event that such a disastrous metadata error occurs before your system is hit by a meteor is that the media files added since your last incremental backup might become inaccessible. Again, remember the specific subject under discussion here.

Another ZFS feature that I’ve always acknowledged as worthwhile *in server environments* is its ease of multi-disk management (which, once again, is pretty irrelevant for most desktop environments, where if multiple fixed disk resources are present at all they’re usually in the form of simple disk-to-disk mirroring where ZFS has little incremental value to offer).

Snapshots are potentially more useful on the desktop, but ZFS is hardly the only desktop file system to offer them: FreeBSD with soft updates does (the fssnap facility seems to do so for other FreeBSD file systems and for pre-ZFS Solaris file systems as well), VxFS does, LVM-level snapshotting appeared in the Linux kernel in 2.6.8, and even Windows Vista (gag) supports an equivalent facility via its ‘previous versions’ feature, for example (the ‘Volume Snapshot Service’ which appeared in XP may be more limited to server environments).

I’m afraid that your enthusiasm for ZFS has made you less than objective. ZFS is no Ferrari, not even a Porsche/Mercedes/BMW (though it might eventually mature to that level of polish and performance): it may be beyond the basic Ford/Chevy/Chrysler arena, but is currently at best comparable to the mid-range Japanese offerings (which are quite respectable, of course, but so are quite a few of ZFS’s competitors – *especially* for the desktop).

- bill