I went diy route, since similar off the shelf product were too expensive and had significant tradeoffs. Availability was also an issue, and I was somewhat cash strapped then. Now I am just cheap within reason.
Primary aim was device similar to HP microserver n40l with more processing power and at least slightly better expansion potential.
Details and photos here:
Only thing I miss is hotswaps capability or at least proper drive cages, but those are impossible to deploy into consumer chassis.
To adress some points you mentioned:
- saturating 1GBE was trivial ten years ago with single drive, and given with any working setup nowaday. Even friggin rasperry pi has enough processing power.
- Do you really need all-flash array? Why? Think about it.
- you seem to need reasonably fast archive and maybe performant scratch space
- have that scratch space on mirrored ssd array and offload archive to raidzX HDD array of whatever level that lets you sleep at night.
- How much storage needs do you have now and how did it develop over time? If you follow up your use pattern, how much storage will you need in 5 or 10 years?
And now some deeper planning:
- ENERGY USE: find out how much do you actually pay for 1 kWH of power.
- Plan around assumption that for each 1W of continous power use:
- consumes 1(W) * 24 (H) * 30 (D) = 720 WH/month = 0,72 kWh/month per watt of baseload
- same applies for yearly calculation, for each W of base power, you will consume at least 0,72 * 12 = 8640 WH/year = 8,64 kWH/year
- now multiply by actual power price and see how much each component or solution will cost you