Good questions.
Here, for example -
iMac 2012 Booting to OS X Utilities & Macintosh HD missing - it's reasonable to assume that the SSD has suddenly failed. No redundancy of data or metadata with the average Fusion Drive hybrid, so - unless there's a backup - the end user may treat the loss of the file system as
catastrophic.
Yes, just as
catastrophic as the failure of the HDD in a single-HDD iMac, or of the SATA SSD or PCIe Flash in a single-drive MacBook, or of any drive in a multi-drive,
non-redundant configuration (like a split SSD/HDD with OS and apps on SSD, data on HDD). How is the failure of a Fusion drive any more catastrophic than any other storage failure in today's lineup of Macs and iOS devices, none of which can be configured for internal RAID? How many thousands of MacRumors threads have similar titles, with no Fusion drive in sight?
To revisit/rephrase Murphy's Law for a moment... Failure is not an option, it's an expectation.
For the most part, redundancy doesn't find its way into consumer products (or even most business products), because the principle purpose of redundancy is sustained up-time, and sustained up-time is not a priority for the vast majority of users. Do you seriously think that the world's PC makers wouldn't be delighted to add $100-$200 to the price of each machine they sell, if there were significant numbers of customers willing to pay for storage redundancy?
With ZFS: if the SSD is given to L2ARC then the file system should be unharmed by failure of the SSD part of the hybrid. And so on (we should probably take this to another topic if you'd like to know more).
I absolutely agree, there are cases where L2ARC (or nearly any other caching scheme but Fusion) will result in a beneficial outcome - should the SSD fail, the system could fail-over to the HDD and keep working (new thread: "Why is my system's performance so much slower all of a sudden?"). There's a more likely failure mode, however... The SSD survives, but the HDD fails. We're back to
catastrophic, aren't we?
In the case of Fusion, L2ARC, a self-contained hybrid drive, or even RAID, a backup is still
essential - drive failure is simply one of many possible hazards, and following a drive failure, unless you have full redundancy, a restore from backup is inevitable.
Anyway, we're talking Apples and Oracles. Note that in that 2008 Oracle L2ARC article, the HDD storage pool consisted of 44
mirrored drives, and the L2ARC cache consisted of 6 SSDs in a RAID-like configuration (read "
Isn't flash memory unreliable? What have you done about that?" in that article, especially the part about successfully yanking out busy L2ARC drives). In that system, a total failure of the L2ARC was highly unlikely. Since nearly every caching scheme but Fusion
is redundant, I doubt they had the kind of discussion we're having here.
Apple approached Fusion with very different design goals. The "competing" design wasn't a cached RAID, it was "OS and apps on SSD, data on HDD." If they were going to charge customers for a 128GB SSD plus 1TB HDD, their customers would expect to get additional, usable storage out of the arrangement. Otherwise, they'd just break the Fusion configuration and use them as separate drives (which some do anyway). The point of Fusion has always been near-SSD performance on a large-capacity volume, at a low, low price. Seriously, if tomorrow, Apple was to offer a machine with a mirrored Fusion drive, how many would actually run it as configured, and how many would break the configuration in order to have additional, independent internal drives?