Freenas + ZFS Build

MetricGnome

Limp Gawd
Joined
Feb 15, 2010
Messages
200
Hi,

I am planning to build a Freenas box using ZFS within the next month. I plan to use this as personal file/media storage. Backup is not important to me, however redundancy/protecting my data is. In the future, I plan to expand this box to serve both my desktop and an HTPC or two.

1) What will you be doing with this PC? Gaming? Photoshop? Web browsing? etc
Personal file/media server

2) What's your budget? Are tax and shipping included?
$600 US

3) Where do you live?
Phoenix, AZ

4) What exact parts do you need for that budget? CPU, RAM, case, etc. Please be very specific.
Everything~

5) If reusing any parts, what parts will you be reusing? Please be especially specific about the power supply. List make and model.
Nope

6) Will you be overclocking?
Nope

7) What size monitor do you have and/or plan to have?
Headless somewhere

8) When do you plan on building/buying the PC?
Within a month

9) What features do you need in a motherboard? RAID? Firewire? Crossfire or SLI support? etc.
Firewire would be nice I suppose, big fan of lots of usb ports, will be using software raid

10) Do you already have a legit and reusable/transferable OS key/license? If yes, what OS? 32bit or 64bit?
Freenas


As I only have $600 US to spend, I'd like to focus the majority of my cash on the most cost effective hard drive count to storage space ratio possible. I plan to expand this box as much as possible in the future and use it for everything with minor upgrades here and there when needed.

My build:
PSU: CORSAIR CMPSU-650TX 650W ATX12V / EPS12V SLI Ready CrossFire Ready 80 PLUS Certified
Memory: G.SKILL 4GB (2 x 2GB) 240-Pin DDR3 SDRAM DDR3 1600 (PC3 12800) Dual Channel Kit
Mobo: Intel BOXDH55HC LGA 1156 Intel H55 HDMI ATX Intel Motherboard
CPU: Intel Core i3-530 Clarkdale 2.93GHz LGA 1156
Case: NZXT WHISPER WHI - 001BK
Drives: Western Digital Caviar Green WD15EARS 1.5TB SATA 3.0Gb/s 3.5" Internal Hard Drive x4

Total before shipping: $984.91

Thanks!
 
Last edited:
No to prescott! Don't touch Pentium 4's for this.

You might want to look into a deal on the i3-530 or i5-650 (or 660). The power consumption will be great. It will necessitate going DDR3, but LGA-1156 has more upgrade options down the road, plus integrated graphics. On special an i3-530 can cost only $30-40 more than the E5300 + that Gigabyte. AMD is good too, but the i3/ i5's are really nice.

Spend $20 and get an Intel Pro/1000 GT NIC, even if it is PCI. Splurging here would be for a PCIe x1 PT or CT card ($35). Tend to get much better performance versus Realtek NICs.

How much storage do you need? If you can swing it, 2x Hitachi 2TB drives can be had for $130 each.
 
Well, I currently only use one 80GB intelx25-m SSD in my rig. Plan is to use this box as a good start to building a media library etc, it'll have it's own dedicated cox business line too. 10/1 I think...

Haven't really considered i3/i5 for this build, seemed a little more pricey then I'd like, but I didn't bother to look.

As redundancy/knowing my shit is safe is my ultimate goal, I'd prefer a 4 hard drives at the minimum, I think. Just not sure what size to get/future proof it....

Should I be considering some sort of PCI SATA card as well?

EDIT: Updated OP with NIC/i3/mobo/ddr3
 
Last edited:
Swap the Gigabyte mobo for an Intel DG43NB. For one thing, the SATA controllers are quite a bit better, and you get a PCIe Intel NIC for 'free'. You also get more SATA, more USB, Firewire, it's made by Intel...

I'd go for the Celeron E3300 or E3400 instead. Cheaper, lower power consumption, and more than adequate performance for a GigE file server.

Might want to go for a bigger PSU if future expandability is important to you, but the 380W should be fine up to 6 disks or so. I'd probably go with a 550-650W for 15 disks or so that you can reasonably fit in a pedestal case.

As far as case, you can either go with something cheap but with decent cooling like this. When you fill the 3.5" slots,this will get you to 9 disks. If you want to go bigger, I like the look of that NZXT.

Those changes (NZXT case, EA650 PSU) bring you to ~$390 (not counting accessories like a CD drive, USB stick to boot from etc.). Doesn't leave a lot for disks I'm afraid. You might need to go the cheap route on the case and PSU if you want to keep in your budget.

As far as disks, you don't have the budget to go past the 6 disks the mobo will support. I'd suggest leaving the SATA card for later when you run out of ports. You can't afford to buy it right now and don't need it yet anyway, and better solutions/better FreeBSD support are constantly coming out.
 
ZFS LOVES RAM. Put as much in the box as you can since ARC will essentially use almost all of it for caching :)

+1 for atleast an i3 also you will want to make sure you run a 64bit distro (another huge benefit for ZFS).

Remember with ZFS you CANNOT do live vdev expansion... the only way it to add more identical vdevs, or replace the drives in your current vdev with higher capacity drives and let it rebuild before swapping the next one.

Pickout a case, and a goal for total data storage, as well as a speed/load estimate (how many users, what type of reads/writes) and than match your layout to match those needs.
 
The Antec EA380 380W PSU is not a good choice considering that once you factor in shipping, it costs just as much as this slightly larger PSU:
$51 - Corsair 400CX 400W PSU

The Corsair has 30A on the +12V rail compared to the EA380's 27A on the +12V rail. Not to mention that it actually comes with a power cord unlike the Antec. So go for the Corsair instead. As an added bonus, there's also a $15 MIR for the Corsair.
 
Thanks guys, really appreciate the help!

Updated OP /w LGA775 Intel set up, ditched the NIC and changed out the PSU for one I personally trust more, plus it's got 8 SATA connectors. :) <3 Corsair!

EDIT: Decided to go with the NZXT case...I really like the look, it's got a pretty nice HD cage configuration I think and even sound dampening foam on the side panel.
 
Should I be considering a HSF/thermal paste for this build? Stock intel HSF should function just fine right? Would really like to be as passive as possible with this thing to reduce noise. Would a passive cooler and some decent airflow work out well for a fileserver?

Thinking of these:

Noctua NH-U9 Dual Copper Heatpipe Cooler

SilverStone Technology SST-NT01 V.2.0 Fanless Cooler with Heatpipes

EDIT: Updated OP *again* After browsing through the WD Green drives on Newegg, it doesn't seem to make much sense to go with anything less then 1TB. I don't need 4 drives, but I *want* 4 drives. I am now OVER budget by a considerable amount. At this price point, I feel like I should be going with LGA 1156 over LGA 775...I am comfortable with being over budget. I said 600, but I really meant I'd be willing to go up to 800 once I got more excited about it. -____________-

EDIT2: Tossed in 4 500GB drives in as well...cache much? :/
 
Last edited:
Stock HSF will be fine, these new chips run pretty cool even with the basic cooling as long as you have decent airflow. If you want to go totally passive something like a Ninja or whatever is en vogue these days should be fine cooling this to reasonable temps at stock/slightly undervolted. I run an E4500 that is quite a bit more power hungry in my HTPC with a fanless Ninja Mini with pretty poor airflow and load temps are about 55C which is not great but acceptable. Definitely add at least a couple good case fans though, hard drives generate a fair amount of heat and you want to keep them happy.

Going LGA1156 would be a good move, but adds to the cost fairly substantially. If you're okay with that, you can't go wrong. Intel's BOXDH55HC looks like a nice option. If you can't tell I'm a big fan of Intel's desktop boards for uber-low-end home servers. They're rock solid, include Intel NICs, and you can pretty much count on things like non-graphics PCIe cards working properly. The competition tends to use shitty Realtek NICs that you need to supplement with a $30 Intel one anyway, so by the time you factor that in, they're cheaper, and you save 1 PCIe 1x slot too.

While we're upping the budget, Tau is right. More RAM would be better.
 
The wife found my post and my budget is back down to $600. :-(

Based upon my application of this box...It'll never be used *that* much, I will torrent on it 24/7, and will use it as the central location for everything that is MetricGnome. That being said....yeah, I'd feel more comfortable with LGA 1156, but I really don't think it's going to matter that much as I highly doubt I will even fully utilize the e3300.
 
Remember with ZFS you CANNOT do live vdev expansion... the only way it to add more identical vdevs, or replace the drives in your current vdev with higher capacity drives and let it rebuild before swapping the next one.

This is an important factor with FreeNAS.

Also remember that this means if you go over the normal 5/6 onboard drive spots, your per drive installation cost will go up by $12.50/ drive (fully populated 8 port supermicro SATA card, the per-port cost goes up if not fully populated). Using $70 1TiB drives it becomes very expensive just to connect them. With Raid-Z/ Raid-Z2, practically speaking you end up adding 3-4 drives minimum everytime you want more capacity.
 
Some recommendations:

  • Make sure you get a 64-bit CPU; not all Celerons may support that.
  • I would suggest a dualcore cpu as minimum, and 4GB RAM as minimum
  • Generally the AMD-platform is better suited since its cheaper; a $50 motherboard may provide you with 6 full-speed SATA and pci-express. Its low power too, though the new Core ix chips are quite good as well.
  • Using the new EARS drives would need special formatting (partitioning) - FreeNAS default way of partitioning may be wrong resulting in lower performance. This is because the new EARS drives are 4KiB sector drives internally, yet still say they are a 512-byte sector drive just like every other HDD and SSD. The EARS drives would be good if you can make sure you partition properly.
  • The 1.5TB disks generally offer best capacity-per-dollar and you would require fewer disks and ease power requirements and physical space as well. It would also keep room to expand without having to use a PCI-express 8-port SATA controller.
  • You can expand your RAID-Z; just not by one disk. You will also be adding a parity drive, so increase the redundancy level.

Expanding your RAID-Z will work like this:
you start with 4 disks in RAID-Z. A year later you get 3 more disks, you execute the command:
zfs add -f existingpool raidz label/drive5 label/drive6 label/drive7

Now your configuration looks like this:

tank
raidz1
label/disk1
label/disk2
label/disk3
label/disk4
raidz1
label/disk5
label/disk6
label/disk7

So now you run two RAID-Z arrays in RAID0 essentially. This adds speed, maintains redundancy since you add another parity disk and would allow copies=2 to make sure both copies land on both RAID-Z arrays.

The downside is that you lose the capacity of the parity disk. So while you added 3 more disks; only the free space of 2 disks is added to your pool's free space.

You don't have to add the same amount of disks each time, but it is recommended. A good setup would be to start with 4 disks, expand with another 4 later on, then another 4 if necessary. That would be best performance-wise. Best of all, your disks do not need to be of the same size. Though you will lose the additional performance if the smaller capacity array is becoming full and ZFS will stop doing RAID0 between both arrays and just use the larger capacity RAID-Z to write to.

So i wouldn't say expanding a RAID-Z is impossible, it's just different than you're used to. It might also make more sense, as you lower the actual redundancy if you add more disks unless you add more parity disks as well. This setup would mean you keep using 1 parity disk per 4 drives; meaning 25% overhead. Still only half of that of mirroring.



He can use cheap PCI-express x1 controllers that add two SATA ports.
 
Make sure you get a 64-bit CPU; not all Celerons may support that.
Anything remotely modern will. Intel hasn't sold a 32-bit only desktop CPU in quite a long time (Early '07 was the last Celeron IIRC). With the most recent ones (Wolfdale) you even get V-T, which is pretty amazing considering a year ago the cheapest CPU with it was nearly $200.
Generally the AMD-platform is better suited since its cheaper; a $50 motherboard may provide you with 6 full-speed SATA and pci-express. Its low power too, though the new Core ix chips are quite good as well.
Do you have experience using them with Solaris/FreeBSD? I've had too many bad experiences with AMD/nVidia desktop chipsets on open-source platforms to recommend them, especially for a server role. But I may just be biased by the previously poor quality of the chipsets on these boards. If you've got some running stable with reasonable performance it's probably not an issue. The budget advantage is small, but real, so probably worth looking into for the OP here. Don't forget to add that Intel PCIe NIC though.

Expanding your RAID-Z will work like this:
This isn't expanding a RAID-Z, it's adding a new one.
 
Last edited:
Anything remotely modern will. Intel hasn't sold a 32-bit only desktop CPU in quite a long time
Alright, nice to hear that. Just some confusion on this part, since both Intel and AMD have opted to disable the 64-bit functionality in some Celeron/Sempron models. As he time goes on these older models may be retired and not sold anymore, but but the confusion is still there.

Do you have experience using them with Solaris/FreeBSD? I've had too many bad experiences with AMD/nVidia desktop chipsets on open-source platforms
I've been using them for a long time. There certainly were crappy chipsets, but that was still in the VIA KT133/266A times; long ago. Since the Athlon 64 things have improved, and generally i prefer the AMD-platform over the Intel-platform mainly due to the available chipsets. AMD may not be into the 1-chip chipset business like nVidia, but their solutions are cheap, stable and low-power. With Micro-ATX boards selling for less than $50, you get yourself a nice AMD760G board perfectly suited for a NAS; onboard light GPU, low power design, six SATA/300 and PCI-express of course.

Older FreeBSD releases also had bad support for the new wave of chipsets; that's all resolved now as FreeBSD has more modern drivers (like the new ahci driver). Especially the onboard NIC didn't always work or work well (such as nVidia onboard NICs). Most of those issues are now resolved in FreeBSD8. So both the hardware has become better, and the support for the hardware in OS like FreeBSD. I can't say anything about OpenSolaris though, though my instincts tell me FreeBSD has wider support of hardware; i could be wrong on this.

As i use FreeBSD8+ZFS13 quite intensively for all my workstations, i can say if there was any instabiliy i would have noticed for sure. That doesn't say it works well for everyone, just that it does in my particular setup. I'm using an AMD quadcore and 8GB RAM and a range of HDDs and SSDs; mostly using RAID-Z.

Don't forget to add that Intel PCIe NIC though.
With FreeBSD8 the onboard NIC should work fine. I would stick to it, unless the onboard doesn't work well. The FreeBSD-drivers for Intel NICs, em, had their own problems in the past, so buying an Intel NIC is no absolute guarantee. Though generally they work well.

This isn't expanding a RAID-Z, it's adding a new one.
Well, it is expanding, essentially from:
1x RAID0 ( RAID-Z 4 disks)
to:
2x RAID0 ( RAID-Z 4 disks, RAID-Z 3 disks)

So you're not expanding the RAID5 part, but putting a new array in RAID0 with the existing one.

How ever you would like to call it, it works. And it's functionality is not that different from what 'expansion' used to be, like on Hardware RAID controllers. The only restriction is that you add one parity disk on every expansion you do, so you only get the additional space when using multiple devices:

expanding with 2 devices = 50% overhead = space only of 1 drive
expanding with 3 devices = 33% overhead = space only of 2 drives
expanding with 4 devices = 25% overhead = space only of 3 drives
.. etc

It might also make more sense, as the more drives you put into an array, the chance on failure becomes higher too. So it makes sense to add another parity drive when expanding.

It does place limits; you can't just buy a new disk whenever you feel like; you have to plan. Start with 4 now, add 4 later, and perhaps another 4. The advantage here is that you're not bound to the capacity of your existing drives. If you uses 1TB in the past, you can get 1.5TB now because they are cheaper-per-GB. That's something you cannot do with expanding RAID5 on hardware RAID; mixing different capacity drives will mean it would treat all disks as being as large as the smallest disk in the array.

So ZFS may have restrictions on expansion, but it's certainly not imposslbe and also has its own advantages over traditional expansion.
 
I'm using an Intel BLKDG45ID G45 Express mATX that I got for under $50 in my FreeNAS. It includes the ICH10R and Intel gigabit NIC onboard, which is nice. It's sort of aimed at being a media center board, so it's got some features that aren't useful for a NAS, but it was what I happened to find on clearance. The biggest drawback seems to be that the x16 slot shares lanes with the onboard graphics, so an addon controller would have to be x1 or PCI, or you'd have to disable the onboard video and use the x16 (as I understand it).

It "just worked" 100% with FreeNAS.


Here are the details on my FreeNAS:
Code:
CPU:	Pentium E2160 Allendale 1MB 65W Dual-Core
Motherboard:	Intel BLKDG45ID G45 Express mATX
Graphics Card:	Intel GMA X4500HD HDMI/DVI
Memory:	2x1GB Crucial LANFest 2K7 DDR2-800
Hard Drive:	1GB MicroSD USB Adapter/ WD10EADS 1TB Green
Power Supply:	OCZ StealthXStream OCZ400SXS 400W
Case:	Enlight 7237
Sound Card:	IDT 92HD73E 7.1
Operating System:	FreeNAS 0.7
CPU Cooling:	Scythe SCSK-1000 Shuriken / GELID FN-PX09-20 92mm

The CPU and RAM were decent deals from FS/FT here. I got the 400W PSU in some deal, $40 I think. I originally bought the Shuriken to steal the fan for my 9800GT/Accelero. The heatsink was just sitting around until I decided to build this FreeNAS. I needed a fan to replace the thin 100mm I stole, and the Gelid was cheap and quiet.

Most of the 5.25" bay covers are missing, and I can't even hear it running unless I put my ear up to the open bays.
 
Certainly a nice board yes! Though it would make it less extensible due to lack of many PCIe lanes. In AMD boards you are able to use the PCI-e x16 slot as i understand it, for a controller. Some BIOS just do not allow booting from a device in that slow, if i recall correctly.
 
ZFS LOVES RAM. Put as much in the box as you can since ARC will essentially use almost all of it for caching :)

+1 for atleast an i3 also you will want to make sure you run a 64bit distro (another huge benefit for ZFS).

Remember with ZFS you CANNOT do live vdev expansion... the only way it to add more identical vdevs, or replace the drives in your current vdev with higher capacity drives and let it rebuild before swapping the next one.

Pickout a case, and a goal for total data storage, as well as a speed/load estimate (how many users, what type of reads/writes) and than match your layout to match those needs.

QFT. Put as much as you can afford in the box. Also -- do NOT run the 32-bit version of freenas.
 
Planning on running 64-bit version, I've added an additional set of my memory listed above for a total of 4GB.

I feel pretty OK overall with my parts now...but I am still agonizing over the hard drives!

"The EARS drives would be good if you can make sure you partition properly."

Define partition properly please, haha. :-P
 
For the most part you wont be partitioning your storage pool in the sense of the word.. you will be breaking it out into datasets (filesystems)

Since ZFS basically has no feasable storage limitations there is no need to 'partition' it (or slice if you will)

your going to want to create one big pool with all your drives, raidz/raidz2/raidz3 and you end up with one large pool (disc size x total discs - parity discs = size) so lets say you have a 5TB pool. than you create a few ZFS data sets in that pool to split your data up essentially. Share them to the windows box with your method of choice (cifs, nfs, samba) and away you go, no need to split that large 5TB pool into anything smaller.

This is how I run the solaris builds I manage... Mind you we are also using cifs (we are actually setting up a test lab right now to compare cifs to nfs)

I would say 4GB RAM should be alright for a home server with low access (1 maybe 2 clients at the same time)

Think about expansion like this, you have a 4 disc Raidz (raid5 but better) with 4x 500GB drives, this gives you 1.5TB usable after parity. Later you want to add more drives. You pickup another 4 500GB drives, and add them as ANOTHER raidz to the same pool as another vdev and you than have a Raid50 style setup (two striped Raid5 arrays) so you gain space, as well as speed, as you need more you add more raidz2 vdevs and your speed increases again.

Most instances when you add more storage your speed requirements are also going to go up... and with 2 arrays you will probobly hit your bottlneck (network speed of the NAS box)

Hmmm I kinda rambled there... but i think i got most of it out conherantly.... need coffee >.>

Planning on running 64-bit version, I've added an additional set of my memory listed above for a total of 4GB.

I feel pretty OK overall with my parts now...but I am still agonizing over the hard drives!

"The EARS drives would be good if you can make sure you partition properly."

Define partition properly please, haha. :-P
 
"The EARS drives would be good if you can make sure you partition properly."

Define partition properly please, haha. :-P
EARS differ from EADS drives because they use so called "Advanced Format" - internally they use 4KiB sectors instead of the usual 512 byte sectors. That means that ideally, you would need to read and write in multiples of 4KiB.

The problem is that standard partitions are created with 63 sectors offset; an old habbit that stuck even though it's a very strange number; 64 would have made sense but not 63. 63 x 512 bytes = 31.5KiB. That's where ZFS will begin working, which means if ZFS writes the first 4KiB it's writing to offset 31.5KiB - 35.5KiB. But since the EARS drives uses internal 4KiB sectors; it would affect 28KiB - 32KiB and 32KiB - 36KiB. It can't just write that 4KiB; it would need to read the contents of two times 4KiB, then process the data, and write the two sectors again.

This is called misalignment, and it is important for RAID arrays, Solid State Drives and new "Advanced Format" HDDs. Normal HDDs are not affected.

The problem is that FreeNAS might partition each individual drive before it is accepted in a software RAID array. That's a design choice of how the FreeNAS GUI works. If it still uses the 63-sector partitioning style, that would create problems when using EARS harddrives.

So to be safe, stick with EADS. The EARS drives are higher reliability though, due to bigger ECC (error correcting) data space.

If you're concerned with this i can test FreeNAS in a VM for you, to see how it works and if it would be a problem when using with EARS drives. Currently i have to finish another project though, so it probably won't be today.
 
@sub.mesa - That would be amazing if you'd take the time to take look into that for me!

@Tau - Thanks for the great explanation on how I would expand this fucker. So, essentially..I create my first raidz, and when I want to add more stuff...I just add another raidz so on and so forth. It's not possible to expand my original raidz and just maintain one?
 
Not at this time, Sun has been working on online cdev expansion for a wile... if I will see it in my lifetime... thats another question :p

You can also 'expand' i use the term expand loosly... by replacing each drive in the vdev with larger capacity drives and than letting it rebuild, and soforth.

some zfs lingo incase you are confused by my terms (or anyone else for that matter)

raidz = single parity like Raid 5
raidz2 = Dual parity like Raid 6
raidz3 = tripple parity
vdev = a group of discs forming one of the above meantioned scemes. 4x discs in Raidz is a vdev.
storagepool = your storage pool made up of the vdevs attached to it.


While it is not as "easy" to expand as hardware based controllers and what not... it does have its advantages... by forcing you to add another vdev you usually double the space and speed... (from 1 vdev to 2).
 
Hmm--Now, can you explain the talk about RAID 0/1 like striping/mirroring with zfs/raidz?

Let's say I get my 4 1TB WD Green EARS drives. What would be my ideal zfs configuration for my first raidz array? I am completely comfortable with upgrading drives in groups of 4. However, can you have say...a raidz array with 4 drives, and another raidz array with 2?
 
Mirroring/striping is just like normal, except no constraints (like onboard controllers not liking 3 way mirrors >.< damnit intel) anyways...

zpool create stripedtank1 c1t1d0 c1t2d0 - Creates a two disc stripe (add more discs to make it a bigger stripe)

zpool create mirroredtank1 mirror c1t1d0 c1t2d0 - creates a two disc mirror (note: mirrored at the DISC level not dataset)

Your ideal configuraton completly depends on your goals with the storage system.

Raidz is basically Raid5 you need a minimum of 3 discs.
Raidz2 is just like raid6 needing a minimum of 4 discs

So based on those requirements you would not be able to add a 2 disc raidz in the future, though you could add a 3+ disc :)

Hmm--Now, can you explain the talk about RAID 0/1 like striping/mirroring with zfs/raidz?

Let's say I get my 4 1TB WD Green EARS drives. What would be my ideal zfs configuration for my first raidz array? I am completely comfortable with upgrading drives in groups of 4. However, can you have say...a raidz array with 4 drives, and another raidz array with 2?
 
So....with my 4 drives, I would create one stripedtank and one mirroredtank and that would be my raidz array?

Or, do you have to decide between striped and mirrored on each array? I'm just starting to get hella confused, lol.
 
No.

there are 5 types of primary Raid parity for ZFS all seperate.

mirroring
striping
Raidz
Raidz2
Raidz3

Then you can put those together... like stripe 2x raidz arrays.

You will most likly setup a 4 drive Raidz, or Raidz2 array.
 
Please remember that one key advantage of ZFS's expand options is that the second array you add can be of higher capacity than the first/initial array. So you can start with 1TB disks, and make the second array 2TB disks because in the future that would be the cheaper-per-GB, for example.

Also, you can create 2-disk RAID5 arrays; there is no reason you can't. In theory you could also make a 1-disk RAID5 array, but that would mean it is degraded; some drivers support this. I'm pretty sure ZFS allows you to create 2-disk RAID-Z arrays. Not that you want to, cause it wouldn't be superior to mirroring, still cost you 50% overhead due to the parity and be slower as well.
 
Ah ha! I think I get it...Hows this sound then...4 1.5TB WD Green drives in raidz. I'll then add 4 more of whatever the next sweet spot is (higher then 2TB) drives in another raidz array and then I have the option to mirror or stripe said raidz arrays, yes?
 
I'm also starting to get a bit curious about an off site backup solution for my NAS. Would anyone have any web hosts/cloud services that are cheap and would work well with freenas? I don't really need automatic backups, I am OK with manual...but would really prefer automatic.

I've worked a little bit with AWS EC2 instances, would that be an option for personal/home use?

EDIT: Maybe this? http://aws.amazon.com/s3/
 
Last edited:
Updated OP with latest build...it's more expensive but I feel a lot more comfortable with the 1156 socket then I do 775 for future upgrades.
 
After some more reading, I'm really loving ZFS....However, being hosed my Samba's slowness is kind of putting a bad taste in my mouth.

FreeNAS supports NFS.....Windows doesn't. But, I found this...http://download.cnet.com/nfsAxe-Win...FS-Server/3000-2085_4-10780016.html?tag=mncol would that allow me to use NFS to access my ZFS?

I'm not convinced there's actually a problem, to be honest. I have saturated GigE both with Samba on Linux and Samba on FreeBSD on various occasions, with very little if any tuning and pretty pedestrian hardware. I'd cross this bridge when you come to it.

Win 7 Enterprise/Ultimate apparently includes an NFS client, and I have used the old Services for UNIX (shouldn't it be Services for Windows? :p) on XP with success, but it was a bit of a bear to set up. Can't offer anything on relative performance though.

As far as cloud backup, you can use S3 via numerous mechanisms, for example the S3 connector for duplicity is interesting. I've not used this particular script before, but I have heard good things (though I don't know of anyone actually using it on S3). There are other options out there too. Most of them are scripts so you shouldn't have any portability issues as long as you can get their environment up and running, which should be fairly well documented for most platforms you might run.

The other option is a service like rsync.net (or something similar, they're the guys I've heard of and seen around) which will let you rsync directly to their servers, and rsync is installed or easy to get virtually anywhere, this should be a very easy solution to get going, but it is a bit expensive (maybe you can find something cheaper, webhosts sometimes offer rsync if you don't mind violating the AUP or finding one that doesn't disallow backup use). As a bonus, they seem to be very generous in their support commitment:

rsync.net FAQ said:
Will you support my use of X ?

Yes. If you are using X with an rsync.net filesystem, we will support and troubleshoot your use of it, regardless of how complex or esoteric that usage or application is. All such support will be handled by a real live Unix engineer, and will never be dealt with by a ticket system, autoresponder, or first level / junior "technician".

Really ?!

Yes. We are not just providing offsite filesystems, we are providing complete, end to end, personal customer support.

Haven't used them either, again just heard good things. Personally I use rdiff-backup to a server I maintain in a colocation facility.

Edit: New build looks great. Thumbs up from me.
 
Hmm, interesting stuff....Can't wait to start buying parts for this thing next week.

I plan to turn this build into a worklog once I get going, will probably cover my freenas install/zfs stuff. :)

Thanks for all the help so far guys!
 
Hmm, thanks for posting that InvisBill.

We really should try and compile a sticky with everything NAS in it. o_O
 
After some more reading, I'm really loving ZFS....However, being hosed my Samba's slowness is kind of putting a bad taste in my mouth.

FreeNAS supports NFS.....Windows doesn't. But, I found this...http://download.cnet.com/nfsAxe-Win...FS-Server/3000-2085_4-10780016.html?tag=mncol would that allow me to use NFS to access my ZFS?

I think the poor samba performance is a FreeNAS issue. Linux ZFS support is poor so not ideal for your use. Maybe OpenSolaris is a good option(?), it has decent ZFS and CIFS support.

I got much better performance with Centos than Freenas on single SATA HD (4GB file to ensure RAM caching doesnt cause too much skew of results):
FreeNAS Write: 44.5MB/s
Centos 5.4 Write: 45MB/S
FreeNAS Read: 45MB/s
Centos 5.4 Read: 70+MB/s
FreeNAS Cpu Usage: 50% (Capped?)
Centos 5.4 CPU Usage: About 20%

Hardware:
Mobile Celeron 410
Intel I3100 based board, SATA with NCQ
Single Seagate 7200.10 HD
Onboard Intel GBE

I know a Mobile Celeron is a weak CPU but it just makes the FreeNAS Samba performance issues more obvious and I think with my hardware raid card and a gigabit 802.3AD pair 200Mbytes/sec should be possible with this little CPU and Centos 5.4.
 
Last edited:
FreeNAS is based on FreeBSD (you seem to be implying its Linux), don't understand why you are quoting performance of a 'single disk' which is pretty close to irrelevant imo...
 
I mention Linux because Centos is Linux based. It seems to me that BSD suffers from poor Samba/CIFS perfromance and Linux has little ZFS support which leaves Opensolaris.

Single disk performance is very relevant, Freenas is already showing CPU limitations with only 1 disk so its unlikely to scale at all with raid, conversely Centos is looking good to scale up for Windows sharing performance.

I'm not interested in ZFS + Win sharing performance myself only Windows sharing performance but wanted to reply to MetricGnomes comment; it is not really Samba that has performance problems, it is the BSD implementation of Samba that seems to have perfromance problems.
 
Last edited:
Back
Top