If you know me my concerns are very much the same as when AsRock Rack started releasing new motherboards after not being in the server board industry for.. 10 years(?). Any company that comes out with something new needs to be tested, vetted, see how the manufacturer handles complaints for problems, etc. Even with the great LSI 2008 products, I don't recommend the 3008 just yet. Not because I don't trust LSI, but there's relatively little data and prior to 9.3.1 virtually everyone had major problems including corrupted pools that ended up going bad before they existed long enough to even store data. (I'm just very conservative and don't like to endorse something until I feel it is safe enough that I would use it myself.)
I'm cautiously optimistic. Marvell has been one of those companies that has regularly made very good products that work great on the platforms they officially support (with a few exceptions of some chipsets of course). They've routinely had poor/no support in Linux and FreeBSD forever-ish(?) so while this is a "good" sign I'd rather wait 2 years before I, personally, start recommending them or consider using them.
This is similar to what I said about AsRock Rack when they came out. Cautiously optimistic, and AsRock Rack has had some problems. Some Marvell controllers still being made in boards like the C2750D4I still are not 100%. Nobody knows why and without a reproduction case (which is being worked on, but hasn't been entirely successful) the issues are not likely to be fixed.
Even now, the P20 firmware on AsRock Rack boards doesn't exist publicly, despite the fact that P20 first came out in October 2014. Now I (and a forum user) have talked to AsRock Rack about the firmware issue, and they are taking action (good on them and quick turnaround too) but it's still a work in progress. So until they actually put out a P20 firmware (ideally 20.00.04 or higher) then they've still missed the mark in my book.
So yeah.. let's see how the Marvell's stand up in the BSD community and how good/bad they actually are when people actually use them because they have them handy and chose to be a potential guinea pig to save money/time/effort over going over more tried-and-true hardware like LSI controllers.
So my opinion, to put it short and sweet is:
Great! Let's talk in 2 years and I'll let you know what the evidence is that I've seen. Until then, the LSI is still king and I won't be recommending the Marvell controllers just yet.
Unfortunately, this may also be self-defeating. If we don't recommend them, we won't necessarily have a large user-base that will actually test them, so we may be saying no in 2 years just because we have little or no data points. :P
We used to see regular zpool corruption when I first started experimenting with ZFS and FreeNAS over 3 years ago. I was pretty freaked out because of how many there were, lack of recovery tools, how quickly and suddenly it can go from "good" to "all data is gone". The occurrences of lost pools are now very few and far between now despite a community of users that is probably 10 times larger than it was 3 years ago, and I attribute much of that to the fact that we now recommend ECC RAM and server-grade hardware. We didn't push for that kind of thing 3 years ago, so we're clearly making good decisions on hardware recommendations and people clearly are using that information more often than not. (yay!)
The problem with bleeding edge is sometimes you bleed to death. I'm not a fan of that. ;)
