I was talking about when you mean to switch the disks (need more capacity, existing ones are too old, ...). Keep in mind, the UBE of Enterprise level SATA is 1E15 (UBE = Unrecoverable Bit Error Rate, and is typically identical to the SAS version of the same disk <same mechanics, so just the controller differs; even the consumer versions share the same mechanics for the same spindle speed & physical size>).
Most of the newer RAID controllers are SAS anyway, as it's cheaper (i.e. current generation controller chips and the one before it, where SAS) since it can run both SAS and SATA disks (just not mix them in the same set).
So under these conditions, there's not much advantage to SAS as previous (faster spindle speeds can increase IOPS, and SAS does have an edge on card to disk communications for both reporting and recovery). But it's not the extreme case it once was.
Your system is important, I get that. But SAS tends to be necessary under very specific conditions that your usage doesn't indicate are actually what you're doing with it.
Hence the mention of SATA (expected this to be the case, as it usually is; the absolute necessity of SAS is quite rare these days).
Even with the additional information, Enterprise grade SATA will serve you just fine, and at a lower cost (keep in mind, you must use enterprise grade disks with a RAID card just for the system to work properly due to the recovery timings programmed into the drive's firmware).
Understandable.
Just keep in mind, that consumer grade SSD's and parity based RAID levels should never be mixed (# of cell writes are too low, so even with wear leveling, the failure rates will be too high due to doubling up on writes; data stripe + parity data every single time the system needs to write any information to the SSD volume). Non parity based levels and consumer grade SSD's can be done, as there are no additional writes due to parity information since it's non-existent.
This is an issue with the card though, not the disk type. A faster controller, even with SATA, would be fine (may take more disks if the IOPS level is extreme), or skipping mechanical and going for SATA based SSD's (best IOPS figures available by a significant margin).
Another thing to consider, is a system's planned lifespan, which is typically ~ 5 years for enterprise grade gear for it's initial use (down-lining systems to less stringent usage can be a way to extend the lifespan and reduce equipment costs per upgrade cycle, but this tend to only be viable in larger corporations).
You definitely need to upgrade your equipment then, as a weekly incremental backup schedule is a complete disaster waiting to happen.
They need to be done at least daily, and full weekly, and more frequently if at all possible. The reason for this, is the longer the time between the cycles, the more work that will need to be re-performed due to falling between the last cycle and time of failure.
You shouldn't be doing this if at all possible, as there's always a strong chance that some work will need to be re-performed, even if a card battery is fitted (holds a sufficient charge).
The reason for this, is that the cache may not be large enough to contain every single bit of processed information waiting to be written to the volume.
This is why you don't want to do such an emergency shut-down unless there is absolutely no way around it (or a PSU dies), and let it complete it's writes. Since loss of power is critical in such a case (assuming the PSU and rest of the system are functional in terms of hardware), the UPS is more critical than the card battery.
Ideally of course, is to have both. But as you've discovered, not all card makers actually make batteries available any longer (poor sales, and UPS systems have become cheaper than they once were).
OK, once the system generates a write, the card then takes that data and computes both the stripes, and the parity data to be written. Once this is complete, it's stored into the card's cache.
Now assuming ALL data to be written fits into the cache, and everything goes out (primary power + UPS <if present>, perhaps even the system's PSU or main board), then the card battery will maintain that data until the battery is depleted.
But this is a big IF, and not statistically the case most of the time in high volume systems. 256 - 512K cache isn't that much these days, especially for larger stripes. Something to think about...
It depends on the models you're looking at. IF for example, you're looking at the 1800/1802 series for Areca, No, you won't find a 4 port card (chip used contains 8 ports, and more are done via an on-board expander chip).
Different reasons (different chips) for ATTO, but the smallest port count is 8 ports as well on the R6xx series.
Both series' from both brands are their fastest cards (SATA 3.0), so they're not cheap. But you may be able to get away with a 1680 series from Areca or an R3xx from ATTO. Only SATA 2.0, but they're far quicker than the pile of crap you're currently using, and better features as well.
I'm not the biggest fan of QNAP, but that's me (I tend to build, as the QNAP is just another computer in a box). I'm not all that impressed with what I've seen of QNAP's reliability or support.If we can get a new card that fixes the VM bottleneck, that would be great, but it's not a requirement. I was thinking more like $450 would be the right price, but I've already spent $700 of time researching it.
Building myself allows better flexibility, reliability (i.e. can even add a proper hardware RAID controller instead of using software & the on-board SATA and/or SAS ports), and at a lower cost, but that's me. I do realize the potential support issues, but I'm assuming that falls on you, and since the budget seems tight, is probably the way to go.
If you're dead set on only 4 ports, then you should look at the ARC-1213-4I. Even adding the extension cable, you'd still be on your target budget (last I checked, the cable was ~$90USD).