Best way to get started?

[jgreco]

I don't think so. When ECC RAM encounters an uncorrectable error, the system is halted.

That'd normally be a double bit error though which is fairly rare.

 

[jgreco]

Unless you have your NAS sitting outside in a field or a tree, the electrical surge is going to have to flow into your facility via some wiring. And it's worth pointing out that there is more than just the power path that an electrical surge can take. I've had a cable modem and the connected router both fried do to a close strike. Which reminds me, I need to go buy that optical ethernet isolator.

... fiber from your demarc to your main switch? ...

I note that Ubiquiti has an EdgeRouter X SFP that can be deployed at the demarc, like $80ish.

https://www.ubnt.com/edgemax/edgerouter-x-sfp/

Just guessin' your home network switch is more of a four figure affair than a two figure affair.

 

[depasseg]

I was looking into an opto-isolator to sit between the cable modem and router, but I'm not sure it's worth the price. That UBNT device looks nice, but the cable modem doesn't have an SFP.

Maybe I'll just go with this:
http://www.amazon.com/dp/B000BKUSS8/?tag=ozlp-20
or
http://www.l-com.com/surge-protecto...00-1000-base-t-cat6-lightning-surge-protector

 

[jgreco]

No, you misunderstand. The UBNT device can be used as a sacrificial device at that price point. Coax Copper -> cable modem -> 1G copper -> $80 UBNT -> 1G fiber -> $$$$witchgear/servers/etc.

Our setup here is not quite that good. Cable and telecom come in to ground block/arrestor as usual. Both sides get handed off to separate APC ProtectNet RM's (each has a dedicated 6ga ground run to building ground), then to CPE, then to another set of independently grounded APC PN RM's, and from there into the 1G/10G aggregation switches that connect to the ethernet jacks all over the place. 4x10GE goes from each of those switches via fiber to core switching which also handles the servers. So in theory a big enough lightning strike can take out a bunch of edge gear...

 

[depasseg]

Ahh, I see. Long week.

I've already got a UBNT ERL. It's disposable, but still takes effort to replace which is why I wanted to put something upstream. I like the looks of the APC PN RM though. Haven't used that before.

 

[jgreco]

On frickin tapatalk here so another halfarsed reply.. I meant to mention that surge suppression PRIOR to the CPE is more important than after; if a surge is successful and hits the CPE then it has TWO routes out the CPE - ethernet AND power. So protect the coax...

http://www.amazon.com/gp/product/B000BMP0XE/?tag=ozlp-20

is the rack mount module.

 

[jgreco]

Ahh, I see. Long week.

I've already got a UBNT ERL. It's disposable, but still takes effort to replace which is why I wanted to put something upstream. I like the looks of the APC PN RM though. Haven't used that before.

The 4-slot chassis modules that are intended to be put into the 1U shelf actually go fairly nicely on a plywood telecom demarc board, so I put four of them up and individually grounded them back to building ground. The modules are, well, modular little circuit boards that slide in and are obviously just some MOV's and a pair of jacks.

I think you know I'm a big believer in fate picking on the unprepared. So I go a bit further than most and it seems to work out well in the end.

 

[rt11]

Well, I ordered the B150 motherboard now. It's not an Intel NIC, and there is onboard RAID (though I won't be enabling it, of course) so we'll see how that goes, but I'm betting Realtek stuff (without any silly QoS features) has good enough support and the RAID hardware will mind its own business. If it's too much of a headache after all I should still be able to use the processor and (ECC) RAM with a server-grade board instead, so I've hedged my bets a little that way. Also found a battery backup/surge protector with an ethernet isolator thingy, so the server should be protected from the elements. Yay paranoia! \o/

Also bought another new WD Green drive identical to the other three, so I have a total of four new WD30EZRX-00D8PB0 disks and two spare WD30EZRX-00MMMB0 disks. The latter are about three years old but they haven't been used too heavily in that time and the SMART report says they're in good shape with about 100k LLC (as they've been running with the default Idle3 timeout) and not a lot of relocated sectors.

So now I'm wondering if I should include the older drives in the pool. It seems highly unlikely to me that I'd lose both of them and one of the new drives at the same time (unless the whole array dies for some other reason than simple disk failure), so I'm thinking RAIDZ2 is a decent option. Is that dumb? What about performance, though... can the i3-6100T deal with the parity calculations for a throughput of at least 130 MB/s?

Oh, and if I do use all six drives, I'd need to start with four and then expand the array afterwards. Cause I have a bunch of data on those old drives an nowhere else to put it while I set up FreeNAS. Are there any (potential) complications in doing that?

And re: unrecoverable memory errors, I missed the point on that one. Of course unrecoverable doesn't mean undetectable, and an unrecoverable error is almost certain to halt the system. I knew that, really. Just forgot that I knew. No error correction in mah brain. Anyway if that study is worth anything then that would be the empirical and statistical evidence I was looking for and, yes, you're not just being overly cautious by recommending ECC memory. It's more or less a requirement unless you can live with something like a 10% risk per year of ZFS silently corrupting your data (and a lower but not negligible risk of it corrupting all your data.)

 

[rt11]

A little update, if anyone's interested, plus a bonus question for extra points..

Turns out the B150m PRO-DH from MSI doesn't, in fact, support ECC memory. The manual says it does, and all resellers of the board repeat that claim, but it's a damned lie. Wouldn't even boot with ECC memory installed. No worries though, incorrect specs so easy refund. Unless I decide to keep the board and CPU. Think it would make a nice HTPC. Dunno.

Anyway, then I got a reasonable deal on an Asrock C2750D4I. Lots of SATA ports, dual intel NIC, and ECC support is given. It seems more or less designed for NAS applications.

Just to get started I did install FreeNAS on a USB stick attached to the MSI board and set up a 4-drive RAIDZ1 pool. Then I moved all my data to that pool. Then I managed to get syncthing running in a jail. And Windows shares, credentials, permissions, ownership and all that, turns out it's a nightmare outside the comfort of a Windows-only network, but everything is finally behaving now. It even recognises my UPS and shuts down nicely 30s into a power failure.

So, new question: Can I simply move the USB stick and the drives to the new motherboard? I'm unclear on how much configuration happens during the initial installation (and why it needs to be done in the first place, as opposed to just distributing FreeNAS as ready-to-run disk image.) I can always start over, but I'd really rather not. :)

 

[Robert Trevellyan]

Can I simply move the USB stick and the drives to the new motherboard?

Yes.

You might have to reconfigure the network settings, but that's all.

 

[TheDubiousDubber]

Yup. Planning to do this myself this weekend when I have time.

As mentioned by Robert you may have to fix some minor things. Your mac address will change so may cause networking issues with the current config, but can easily be fixed. In addition, if you are changing to a new HBA or something of the sort it can also cause issue as a result of different firmware, etc. However, if you're using onboard SATA ports then you don't really have to worry about that.

 

[jgreco]

Be aware that the Marvell SATA ports on the c2750d4i are not necessarily well-supported and may give you significant trouble on FreeNAS. You can safely use the other ports though.

 

[rt11]

A little update for anyone interested:

The C2750D4I is weird. First time I turned it on it just sat there with a blinking LED, no POST, no VGA output at all. Thought maybe the RAM was defective so I swapped out some regular DDR3 i knew to be working. Still wouldn't boot. Eventually I gave up and went to make coffee. I wouldn't have thought to leave it on for five minutes, but that just happened to be how I left it, and when I came back the monitor was flickering on and off, with alternating messages of "no signal" and "unsupported signal", but a garbled FreeNAS boot sequence half visible in the background. Eventually the image stabilised, though, and FreeNAS was up and running. So I swapped in the ECC RAM and it took several minutes to start up again, but it did boot eventually.

Rebooting is quick enough, but it still goes through the same thing whenever i cold boot. Weird. Maybe it runs a long self diagnostic or something, who knows. But anyway, for anyone having problems with that board, I recommend giving it a good long time to start up, because apparently it needs it.

With the board running, FreeNAS appears to be very stable. The pool is present and it retained all of my settings including the Syncthing jail. Even network settings were retained. Only thing I needed to reconfigure was the USB port for the BBU which took all of five seconds. Pretty impressive, TBH.

So about the Marvell SATA controller:

I plan to add another 4-drive vdev some time in the future. I've only got two ports left on the intel controller, though. So can anyone elaborate on the problems with the Marvell SE9230 controller? And if it's really that problematic, who can recommend a suitable expansion board? (I have one PCIe 2.0 x 8 slot.)

 

[TheDubiousDubber]

Glad its working for you so far. I can't elaborate on problems with the Marvell controller other than knowing that others have reported problems in the past. As a result it is standard practice to avoid them which in turn means if you do have problems with them it is unlikely you will find anyone who can offer up much help. As far as adding more ports, most people recommend the M1015 HBA flashed to IT mode along with some breakout cables.

 

[Ericloewe]

A little update for anyone interested:

The C2750D4I is weird. First time I turned it on it just sat there with a blinking LED, no POST, no VGA output at all. Thought maybe the RAM was defective so I swapped out some regular DDR3 i knew to be working. Still wouldn't boot. Eventually I gave up and went to make coffee. I wouldn't have thought to leave it on for five minutes, but that just happened to be how I left it, and when I came back the monitor was flickering on and off, with alternating messages of "no signal" and "unsupported signal", but a garbled FreeNAS boot sequence half visible in the background. Eventually the image stabilised, though, and FreeNAS was up and running. So I swapped in the ECC RAM and it took several minutes to start up again, but it did boot eventually.

Rebooting is quick enough, but it still goes through the same thing whenever i cold boot. Weird. Maybe it runs a long self diagnostic or something, who knows. But anyway, for anyone having problems with that board, I recommend giving it a good long time to start up, because apparently it needs it.

With the board running, FreeNAS appears to be very stable. The pool is present and it retained all of my settings including the Syncthing jail. Even network settings were retained. Only thing I needed to reconfigure was the USB port for the BBU which took all of five seconds. Pretty impressive, TBH.

So about the Marvell SATA controller:

I plan to add another 4-drive vdev some time in the future. I've only got two ports left on the intel controller, though. So can anyone elaborate on the problems with the Marvell SE9230 controller? And if it's really that problematic, who can recommend a suitable expansion board? (I have one PCIe 2.0 x 8 slot.)

Expect something between "constant errors" and "works reliably, but is very slow".

The only reliable PCI-e SATA controllers are LSI's SAS2 controllers.