FreeNAS as backend for Mac server

[dtemp]

I'm growing tired of the Promise Pegasus2 R8 DAS on my Mac server that is the storage for File Sharing and Time Machine services. Wondering how I might use FreeNAS instead as a "DAS" on the Mac.

For complicated reasons, it seems like I still need the Mac to provide the actual services/authentication, so using a standalone FreeNAS box seems like it isn't in the cards.

Basically the options are:
1) Running 1GbE between the FreeNAS and Mac (their own backend network), using a NFS share, and having Time Machine and File Sharing services point to the share. (Do those services even allow pointing to a network share?)
2) Running 1GbE between them and doing iSCSI with some third-party Mac iSCSI initiator I assume
3) Getting a thunderbolt-to-10GbE SFP+ adapter and connecting the units with a twinax cable, and using either NFS/iSCSI. Not sure if this product exists that accepts twinax.

Any thoughts?

 

[Mlovelace]

Are you doing file sharing and TM service through OS X server? You can setup individual AFP and TM shares directly on the freeNAS box and associate them with specific user accounts (taking OS X server out of the loop for that).

All three of the options you cited above won't be as fast as the DAS thunderbolt2 enclosure, but I don't know what your needs are (more then 8 disks?).

EDIT: ATTO makes a 10gbe sfp+ to thunderbolt device http://www.attotech.com/products/product.php?scat=32&sku=TLNS-1101-D00

 

[dtemp]

Yeah theres no doubt that the interfaces will be slower... but I don't think I need it. Clients will be accessing the server via a single 1GbE link via heavy protocols (SMB/AFP), so having 1GbE NFS/iSCSI on the backend should be fine. I can tell you that $1K+ 10GbE adapter is definitely not in the budget.

What I need is reliability. This Promise unit has a clunky native Mac app that frequently is unresponsive or otherwise causing issues. The whole unit was unresponsive during a single HDD failure in a RAID6, requiring a reboot, which isn't acceptable. I've got a support ticket open with Promise now, and unless something actually gets fixed/replaced, it's getting eBay'd. I'm familiar with FreeNAS and it's failure modes; I've done the whole thing where you replace drives one at a time to grow a ZFS array, I've had drives organically die, I even tried setting up ZFS once with only 4GB of RAM and watched the spectacular failure (everyone says 8GB is the minimum for a reason folks!).

I'd like to avoid setting up a ZFS dataset/quota for each user and having to set up directory services so they can sign on the FreeNAS box with their Mac server password.

The big unknown here is if having dozens of people backing up at once is going to choke over NFS or not. As far as FreeNAS/Mac is concerned, the clients will be opening .sparsebundle files with a zillion "bands" inside, and a lot of random IO while it prepares incremental backups. I posted a thread a while ago asking if anyone uses FreeNAS for an office with 50 people backing up every hour, and I got crickets.

 

[cyberjock]

Two things:

1. If you plan to go with 50 people doing backups, you'd better have a 10Gb backbone from your FreeNAS server to the network. That's going to be serious throughput and, for a company that can afford 50 Apple devices, the cost of some 10Gb equipment to ensure good backups is not asking for the stars.
2. I've seen someone use one of those Atto 10Gb cards. They are interesting, but they also create a boatload of unknowns. It's proprietary hardware with a proprietary conversion interface. We got good performance with real-world tests, but their built-in benchmark test always showed numbers that were less than 1/5th of what we expected (and what real-world showed). No doubt that once again if you aren't intimately familiar with a given benchmark software you shouldn't be using it.

 

[dtemp]

What if I told you that all 50 of these devices are on WiFi connecting to two APs... so having aggregate bandwidth above around 1.5Gbps isn't going to happen anyway! :)

Yeah I know... there should probably be a couple more APs... the contract for the wireless networking guy is at legal review. One thing that definitely isn't going to happen is having any of these machines backup directly over wired ethernet.

 

[cyberjock]

50 devices on wi-fi are a fail. It doesn't matter if "FreeNAS can handle the workload". Clearly your network has between no-chance and 0%-chance of handling that. You're getting ahead of the horse by asking about FreeNAS with a network like that.

And the number of APs won't help as much as you want. One laptop1 talks to AP1 and laptop2-30 have to work around that "interference" between AP2 and AP8, you'll find that you bottlenecked yourself because of the limited frequencies available for signals.

So I think you guys need to *seriously* go back to the drawing board with wired ethernet because without it, you are pretty much going to be very disappointed with the final product.

Can I guess that a "manager" at your company decided this was a "great idea"?

 

[mjws00]

Nah. He's in a high rise with a bunch of roaming mac laptops. So wired isn't in the cards. We were the crickets in his other thread ;)

 

[dtemp]

Beat me to it :) I'd be the first to claim it is far from ideal. But we work with what we can.

 

[cyberjock]

I don't think there is such a thing as "working with this" with the scale you are requiring.

 

[mjws00]

The only way to answer this question reliably is to test, imho. You can download the iscsi initiator trial (14day). You need to ensure the server.app can use iscsi as valid storage for time machine. Throw up a FreeNAS target, and see how it does. We know it could do 1 user, probably 10, maybe 50. Incoming bandwidth will always be small... so we don't have to worry about the backend choking. If I had to solve this problem, I'd literally grab my little e3 with 32GB, take it to work, throw a couple mirrored drives in it and try it. If we can scale to 10 users and measure we can estimate our needs and bottlenecks from there.

No idea how much hardware the "mac server" will need, but that mini was already choking iirc. The other concern I'd have is latency. That thunderbolt array is very very fast compared to 1Gbe. Time Machine does a ton of excess prep and checking, and it is hard to guess at how that behaves over the wire at scale.

You may be able to scale-out at that point both to increase available processing and bandwidth. Mini's are cheap, and a decent FN box can serve up multiple safe targets.

Two bits. I'd want to break it down and find limits, then split loads as necessary. Or just buy an enterprise backup solution. :)

 

[Mlovelace]

Are you running the server.app on the apple "server" mini? FWIW you can run OS X as a vm in ESXi, you have to edit the host but it's easily done.:) That would make servicing the 50 clients a bit more manageable.