Most efficient build

[cazten]

Hey guys,

New to the site here. Been reading alot but am looking for some advise to point me in the right direction for my slim requirements.

Im looking to build a NAS to host my media collection. Pretty much everything is large scale Bluray rips or MKV Remux in the 10-40gb range per movie.

My requirements for the system is simple.

1.Be able to transfer at 50+Mbits/s to feed my XBMC box stutter free
2. Be able to run a bit torrent client in the backround 24/7, without effecting streaming performance.

3. While 50mbit is as low as I can possibly go (which is pretty stupid low), Id like to milk as high of speeds as I can get while meeting my hardware requirements.
** As a side note, im pretty money bound. Ill likely be starting off with just 1 HDD, and expanding as time/money requires. So far my media collection is just 250GB. This solution is in the works as thats my entire laptops Hard drive lol.


So onto hardware...
As a 24/7 bit torrent server, I want as absolute high energy efficiency as I can get.
I know im looking mini ITX. The question is which Mini ITX mobo/cpu combo is the best energy efficient solution on the boards right now fully supported. AMD-C60 combo? Does it get any better than that? Is c-60 already crossing the barrier speeds being far to slow?

 

[cyberjock]

I'd recommend doing a little more homework...

You'll notice:
1. You cannot easily start with a small zpool size and then "expand" it easily. Read my guide. Your options really are build it exactly how you want it with all of the hard drive space you plan to use up until a future point, or abandon ZFS. If you had read the sticky in the noob section you'd have figured this out....
2. If you are "pretty money bound".. see #1.
3. Torrents can bring zpools to their knees depending on how fast your internet is. Some people have had zpools go unresponsive.
4. If you had done your homework you'd have realized that asking your question always yields the same general answers.

So yeah, you need to do a lot more homework before you start trying to pick out your hardware. And if you think someone is going to just give you the answer, think again. The FreeNAS project(and FreeBSD in particular) leaves the administrator with the hard decisions in terms of how much money to spend, how to spend it, how to be as efficient as you are willing to pay, etc. There is no canned response, not even a somewhat easy canned response. If there was then it would be a forum sticky and all the people that show up and ask the same question as you would not have to ask the question.

There is a section for builds of FreeNAS where people posted the parts they used. You should go there unless you plan to do ALL of the homework yourself. I will tell you that anyone that says "I don't have alot of money" will almost certainly not go with FreeNAS. The initial cost is too high for many people.

 

[Stephens]

New to the site here. Been reading alot but am looking for some advise to point me in the right direction for my slim requirements.

Everything in your post has been covered here recently, including this "incremental build" you want to do. FreeNAS isn't the best solution for that. The AMD C-60 has also been covered (several times). Why not just make a Windows server?

 

[jgreco]

The initial cost is too high for many people.

The problem is, if you want efficient, it's hard to build a truly efficient FreeNAS server that's small.

Consider these little NASpliances made by your Seagates, Netgears, iomega's, etc. Take a SoC, stick it in with a Busybox of some sort, instant Linux based proprietary NAS - but one that takes maybe 5 watts. Effectively untouchable by FreeNAS, partly because you can't find an Intel platform that efficient, partly because ZFS wants its 8GB or whatever. I keep seeing the iomega ix2-dl around for $80-$100 every few months. 5 watts. Gigabit ethernet capable of pushing 50-60MB/sec. Hard to argue with.

For a slightly larger, more powerful system, the vendors move up to Atom. FreeNAS can compete in that arena, but ZFS is still a heavy toll, and results in a potentially slower system than if you used UFS. Costs can come out similarly though. For all practical purposes, the low-power AMD stuff and HP N40L are basically in this same general class.

Where FreeNAS gains the edge is large systems. Look at the price incremental for the vendor NAS solutions once you get out to more than four drives, and it's kind of ridiculous. The Thecus N4100EVO is at NewEgg for $255 right now, but beyond that you're very quickly up in the thousand dollar range. You can pretty easily build a six drive FreeNAS unit for much less, and as you get out to a dozen drives, the vendor solutions are laughable AND pricey.

 

[CAlbertson]

Other have been saying "don't bother with freeNAS and ZFS, for only 250GB of data. I agree unless the goal is education.

The most practical thing for such a small scale NAS is to use an old notebook PC and share the internal disk over the network. The notebook has it's own built-in battery and screen and takes very little space. This can work for up to a few TB of data. Do not use ZFS for this.

That said I just set up a small FreeNAS server using just one 500GB 2.5 inch notebook drive for data. Performance is poor but the goal here was to learn about freeNAS. Then later replace the entire server. So I'd say to just build using whatever free/junk PC hardware you happen to have. Then do the whole thing over again later.

But practically speaking wait until you need to store more data. a freeNAS server is not worth the effort unless you have several terrabytes of data. Linux on a notebook will server your 250GB just fine. Run a freeNAS inside a virtual environment (like vmware) to learn about it.

 

[jgreco]

Why wouldn't you use FreeNAS and UFS for such a task?

 

[FireWire2]

Not a perfect solution - but it woks

I have used both ITX MB with Realtek NIC and Intel NIC.
I would pay a bit extra for Intel NIC at ANYTIME. You can see vast diff between them...

There is no perfect solution for your need, but I think there is a solution:

Here is my suggestion (note: ProBuilder! Please adding your input)

mother board - ITX
http://www.amazon.com/dp/B007BHTMX6/?tag=ozlp-20
http://www.amazon.com/dp/B007MS9OI2/?tag=ozlp-20

2GB RAM

Now you can built your single drive NAS... Here is the kicker....

Add more drives - this time you have to add 5 drives as once
RAID controller - SPM393-I http://www.amazon.com/dp/B004JPUZWU/?tag=ozlp-20
With 5x GREEN drives (low power) - create a SINGLE RAID5 volume

Connected to the 2nd SATA port... FreeNAS will see it as a SINGLE HDD, but HUGE. Protected single HDD.
Move data over
Decommission the singe HDD

SPM393-i is a lower power hardware raid controller, it DOES NOT requires ANY drivers, as long as it connects to a SATA port. It works.

Here is my system base on similar raid controller - It'd been in service close to 3yrs now
http://www.tomshardware.com/forum/265641-32-40tb-server-performance-issue

 

[CAlbertson]

Why wouldn't you use FreeNAS and UFS for such a task?

I tested an Atom powered desktop system with 7200RPM Seagate barracuda drives used UFS and got about 4 megabytes per second max. Then I simply shared a drive in a Macbook Pro and got 62 megabytes per second. Both Mac OS X and FreeNAS are BSD UNIX based but the Macbook is a dramatically better file server than the Atom based desktop. (it also costs 4X more) It is likely faster because of (1) the i5 processor s much faster and (2) Apple spends time to hand tune the the kernel level drivers for the hardware.

I think the minimum usable FreeNAS is an Intel i3 at about 2.8 GHz with about 8GB or RAM and then your disks.

 

[RvdKraats]

Not sure if this will help you in any way, but my setup is a HP XW4600 (Core 2 Quad), with 4 disks and 8GB memory. I didn't bother to check the power consumption (got it for free as it was written off by my company), but performance-wise it's more than enough to stream (by wire) HD movies to my Boxee through my old 100Mbit switch. CPU graphs hardly register any activity at all.

 

[CAlbertson]

Not sure if this will help you in any way, but my setup is a HP XW4600 (Core 2 Quad), with 4 disks and 8GB memory. I didn't bother to check the power consumption (got it for free as it was written off by my company), but performance-wise it's more than enough to stream (by wire) HD movies to my Boxee through my old 100Mbit switch. CPU graphs hardly register any activity at all.

If it is using a 100Mbit switch the performance is capped less then 1/2 what you can get from a USB 2 external drive. Possibly the server could do far better but that switch is a big-time bottle neck. Put a watt meter on the old HP and look at your local cost of power. I found in some cases it is cheaper to pass on the "free" computer especially if you use AC in summer. With AC ever watt used by the computer costs you more than double in AC power consumption.

All that said, while the Atom is good on energy it's performance is very poor. I am beginning to thing that FreeNAS can not run on an energy efficient platform.

 

[jgreco]

Of course FreeNAS can run on an energy efficient platform, but there are lower power energy efficient platforms available that FreeNAS cannot run on. Consider, for example, something like iomega's ix2-dl, which runs about 6-8 watts with SSD's or laptop hard drives. These platforms are built around various SoC implementations, but they are also limited in performance, and number of drives supported, etc.

FreeNAS with ZFS isn't really intended to run on a really small platform, but we appear to live in an age when 8GB is no longer "large memory."

As for why you would only get 4MB/sec out of an Atom with UFS, ?????... I have an Atom D525 on the bench right now with some Barracuda 400's (peak out around 60-70MB/sec). I stuck an XP VM on the same vlan. Created two filesystems on two disks on the NAS (because the VM datastores for lab are pretty slow and would be limiting). I am currently watching the XP VM copy a large file at 15MB/sec, which means the Atom is simultaneously reading AND writing that data (30MB/sec throughput). It's a vanilla FreeNAS install, no tuning, no tweaks, nothing other than create filesystems, create shares, go. The Atom is showing 80% idle, and it is very likely that network clutter between the bench network and the VM lab network is reducing throughput somewhat.

All things considered, I think I'm not too impressed with the Atoms, the energy consumption of a Xeon E3-1230 doing similar work is only a dozen watts more, and it has the potential to do so much more if needed.

 

[CAlbertson]

Any way I can debug my setup? I'd be happy with 15MB/Sec but I'm not seeing even 4MB/Sec. It is not the network, I plug in a Macbook Pro and share a directory and go 60MB/Sec using the same network cable.

It might be the Realtek controller on this motherboard. My next step is to install Linux. If it also is limited to 4MB/Sec then I'll blame the hardware.

I'd like to understand this better and find the bottle neck. What should I look at?

...
As for why you would only get 4MB/sec out of an Atom with UFS, ?????... I have an Atom D525 on the bench right now with some Barracuda 400's (peak out around 60-70MB/sec). I stuck an XP VM on the same vlan. Created two filesystems on two disks on the NAS (because the VM datastores for lab are pretty slow and would be limiting). I am currently watching the XP VM copy a large file at 15MB/sec, which means the Atom is simultaneously reading AND writing that data (30MB/sec throughput). It's a vanilla FreeNAS install, no tuning, no tweaks, nothing other than create filesystems, create shares, go. The Atom is showing 80% idle, and it is very likely that network clutter between the bench network and the VM lab network is reducing throughput somewhat.

 

[jgreco]

Any way I can debug my setup? I'd be happy with 15MB/Sec but I'm not seeing even 4MB/Sec. It is not the network, I plug in a Macbook Pro and share a directory and go 60MB/Sec using the same network cable.

It might be the Realtek controller on this motherboard. My next step is to install Linux. If it also is limited to 4MB/Sec then I'll blame the hardware.

I'd like to understand this better and find the bottle neck. What should I look at?

Everything. But do it one thing at a time. "It is not the network"? Well, actually test that. Putting a different set of tires on the car and then racing is not proof that you don't have a flat tire. Test in-place. Use the tools. Set up iperf on the NAS and on a client that you've seen to have problems, and see what kind of speeds you get. Have you double-checked that your network interface is configured correctly? If you do an "ifconfig rl0" (or whatever type of interface it has), does it show 1000/full as negotiated?

How fast is the disk itself? I've seen failing disks that sort-of-work but respond slowly. If you do a "dd if=/dev/zero of=file bs=1048576" on your datastore, how fast can it write? (do a few control-T's and it'll report during the test). Note that this incantation will fill your disk if left to run.

And don't forget a NAS is an abstraction layer of sorts. That means that overall performance isn't going to necessarily be what you measure out of individual components of the NAS's subsystems. For example, that NAS on the bench here with 60MB/sec disks was only CIFS'ing at around 15MB/sec. It could probably be made to go faster than the 15 but it can't possibly go faster than 60, and I'd say even 30 would be lucky.

 

[CAlbertson]

Everything. But do it one thing at a time. "It is not the network"? Well, actually test that. Putting a different set of tires on the car and then racing is not proof that you don't have a flat tire. Test in-place. Use the tools. Set up iperf on the NAS and on a client that you've seen to have problems, and see what kind of speeds you get. Have you double-checked that your network interface is configured correctly? If you do an "ifconfig rl0" (or whatever type of interface it has), does it show 1000/full as negotiated?

How fast is the disk itself? I've seen failing disks that sort-of-work but respond slowly. If you do a "dd if=/dev/zero of=file bs=1048576" on your datastore, how fast can it write? (do a few control-T's and it'll report during the test). Note that this incantation will fill your disk if left to run.

And don't forget a NAS is an abstraction layer of sorts. That means that overall performance isn't going to necessarily be what you measure out of individual components of the NAS's subsystems. For example, that NAS on the bench here with 60MB/sec disks was only CIFS'ing at around 15MB/sec. It could probably be made to go faster than the 15 but it can't possibly go faster than 60, and I'd say even 30 would be lucky.

OK, I think I found the bottle neck but how to corect it?

I plugged three computers (two Macs and one FreeNAS) into the same 1000BaseT switch. All three LEDS on the switch indicate "1000BaseT" speeds. I verify this by running "ifconfig" on each computer and it shows each has auto selected 1000BaseT media, So far so good.

Step 1. Verify disk write speed on FreeNAS using "dd". I see 109 MBytes/sec. This is "good enough"

Step 2. To verify the Macs I run iperf on each of them and see a solid 930000 or better Mbits/sec. I test each Mac as both a client and server iperf and the result is the same.

Step 3. Run iperf server on FreeNAS. Results are very poor. The average over 10 seconds is about 40 MBits per second but the numbers move every second from a low of 8 to a high of 116 but mostly stay around mid two digits numbers of Mbits/sec This is not good.

Step 4. Run iperf client on FreeNAS. Results are decent with 690 Mbits/sec. Not great performance but decent.

Conclusion my FreeNAS system's Ethernet performance is very asymmetric

Question: What to check next? OK I'm guessing, Interrupt problem in Ethernet port? I don't know how to verify this guess.

 

[jgreco]

Make sure it's not only 1000baseT, but 1000baseT and full-duplex and autoselect. It's not really supposed to be an issue in this modern age but you may have dodgy hardware of some sort, and if some of your links are not configured correctly, ... problem!

Hard to know what to do if it's the card. PC's are finicky beasts and you can play look-around-the-BIOS to see what might be affecting it (if anything). Or it could be crappy hardware and that's just life. You appear to be looking to replace the interface which is pretty much the least frustrating first step.