OK, maybe I wasn´t very clear.... I don´t mean I´m going to move things ´dynamically´ between disks, just that I decide what´s important enough to keep on the SSD and what´s not (if necessary, one can always make a symbolic link to something on the HD if it´s necessary to ´pretend´ it´s on the system disk). For example, if I were to keep Garage Band at all, I would put it, and it´s very substantial Application Support folders on the HD, likewise all of my raw image files except possible the ones I´m still working actively with. The Lightroom data base and previews, however, would stay on the SSD.
In fact, this is exactly what I´m doing right now on my old MP, which does have an SSD as system disk, and a 1 TB HD for the rest (the other 2 bays are for backup purposes). With the Mini, I´ll use a 1 TB HD in a FW 800 enclosure for the same ´rest´; remember that the HD part of the Fusion drive is a 5400 rpm one.
Finally, my conclusion is obviously not based on personal experience with a Fusion drive, but mainly on this Digilloyd article, along with similar ones from ArsTechnica. In the linked article, Lloyd specifically states that
"What I have NOT been able to find is any intelligent migration activity: repeated viewing of images which are known to be on the hard drive does NOT cause migration to the SSD. Hence performance remains poor for reads on files that I’ve used for real-world viewing over and over.
Hence storing a Lightroom catalog or similar on a Fusion drive is massively inferior to storing it on an explicit SSD volume. Where people get fooled is starting to use a Fusion drive which has an SSD still with ample space. That’s not Fusion at work, that’s just the SSD not yet full."
That just about settles it for me...
(And, Elliot, I´ll return with my experience when I´ve got the delivery and get up and running again...)
With all the respect to you and Mr. Chambers, that review (maybe more than the Fusion technology) is mediocre at best.
The author writes that even opening an image a number of times the data is kept on the HDD. How many times did he try? We don't know.
Who tells that after how many times a block is moved? Who tells wether it's the same no matter what the size of the read data?
Any serious test clearly demonstrated that reading many times the same file (maybe even just one more than what Mr. Chambers did), it (or some blocks) get moved indeed.
I don't have any Fusion Drive test, I only spoke with computer engineer colleagues that have one.
It would be questionable to move data after a small number of reads. That would stress the mechanics over necessary.
Other sources:Fusion Drive: An OverviewPushing a Fusion Drive to its limits
As for you example, you may leave very small portions of GarageBand on the SSD, or none, depending on your usage. Remember: blocks are fragmented at MB sizes, not GB. It's a very fine granularity.
I disagree with the author regarding the Lightroom Catalog. My one (and most probably yours too) gets modified -several times- a day. It's the most common candidate to be left on the SSD. Being the catalog a SQLite DB, in the worst case it may leave old portions which weren't accessed since a long time, yet I think that most of it will always remain on the SSD.
Regarding the 5400rpm disk, benchmarks show about 80 MBps in sustained write speed. Even FW800 may struggle to reach those speeds. Definitely not a bottleneck.
The dual approach has a single advantage over the Fusion. You may decide what to keep where no matter what. However, since you'll most probably decided basing on performance, you'll end to do what Fusion does for you under the hood, with much less flexibility. Is it worth it? Maybe, or maybe not.