The small Atom systems with different configurations are close to what you describe. And indeed, you get much better insight in to what the performance of the pieces are when you can take your time.
I did look at the link you provided. I made some changes to the rc.d file that brings up the mysql SSD database without much luck. There was some flakeyness with the server port being up but the mysql client couldn't connect, which is what finally drove me to just "delete-reInstall". Don't like it, but. My core background is hardware and firmware, started with Unix in '76 at Purdue. (and a PDP-8 in 1972). I only wish my testing in software was as good as it is in hardware and embedded software...
The SSDs do "wear leveling". This spreads the write cycles around on the flash memory evenly. An MLC-2 can write a given flash bank about 1200 times. The MLC-2 means 2 bits per cell stored as an analog voltage. Most SSDs in consumer stuff (and perhaps in the big SSD long term storage devices) are MLC-3, which is 3 bits per cell and can write a flash bank about 300 times. Erasing and writing the cells causes the degradation that limits the number of write cycles you can do. The new drives may also have abilities to re-write blocks if they have not been written in a while, but this is some new stuff. It's a concern for me because I'm going for 20 year life. Not so much for a laptop or desktop that will be relaced in 4 to 8 years for most people. The 'trim" capacities in FreeBsd are also a big help. It tells the SSD which blocks are no longer in use by the OS. This maximizes the effectiveness of wear leveling.
For decades the spinning rust was a dream device. You can read and write forever, you just need to replace them eventually. Kinda like a well pump- it will pump forever but only start so many times. So the no moving parts of SSDs, their zero latencies, physical ruggedness and size are a win but it's taking some time for the rest of a computer systems to adjust how things are done so they can really take advantage of SSDs. There are people working on storage that is fast and can be read/write forever, time will tell. In my embedded work I really like the small processors with MRAM (magnetic memory, reads are destructive, just like core "everything old is new again".) But those are small < 65 KB devices. They are wickedly fun to program and optimize.
The total storage on the 4TB of SSD in the system right now is about 5GB. Of course that will grow, but even if it reached 500GB the number of writes would be about 1200 x 8 or 10K writes per block. Not bad. At 50GB that goes to 100K writes of all your data.
Note also for this mobile use, the lower power of the SSD (especially since they are idle 99+% of the time) is a big win too when compared to spinning rust. The deployable servers draw about 25 watts most of the time, and can go to about 50W if all the cores are really cooking, memory is changing and the SSDs are storying stuff.
I hope to write up what I'm doing both with the "database in a RamDisk" and the "mobile server" stuff. IDK if there's a way to submit that to this forum or not. Hopefully someday. Just like I can look at a schematic and always ask good questions, I'm sure same is true for software. Especially mine. It looks like an embedded C programmer trapped in an object orient environment. Todays decisions not withstanding, at least most of the time I do OK on testing.
Wrote a lot here, hope that's OK....
