ARM port of FreeNAS planned??

Do you think FreeNAS should have an ARM64 port

  • YES

    Votes: 13 40.6%
  • NO

    Votes: 19 59.4%

  • Total voters
    32
Status
Not open for further replies.

Bernard Mentink

Contributor
Joined
Apr 2, 2016
Messages
193
I ask this question because there is more and more push for ARM servers, in particular powerful 64bit ARM servers ..

The reason for this is of course the much better power/performance factor than x86 (at least at the moment).

Power bills for 24/7 server farms are horrendous .. millions of dollars can be saved every year (and the planet) by this sort of migration ..

FreeBSD already has a port for ARM64 and in particular on some reference server hardware.
https://www.freebsd.org/news/status/report-2015-10-2015-12.html#FreeBSD/arm64
https://www.freebsd.org/news/status/report-2015-07-2015-09.html#FreeBSD-on-Cavium-ThunderX-(arm64)

It seems to make sense that FreeNAS adopt this for the new hardware that is coming out ..

Does FreeNAS have a plan to address this in the future ... (being of course the best server OS out there ..):p;);)
 
Last edited:

Bernard Mentink

Contributor
Joined
Apr 2, 2016
Messages
193
Last edited:

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
Does FreeNAS have a plan to address this in the future ... (being of course the best server OS out there ..):p;);)

FreeNAS definitely isn't "the best server OS out there", though if you want to say it's the best NAS platform out there, feel free. As a server platform, FreeNAS is big, fat, and eats too many resources.

As for an ARM port, it'd be nice to see, but with limited developer resources, it doesn't seem that likely. Further, ZFS specializes in torturing the CPU for things like RAIDZ calculations, so an ARM version might be very much less powerful than you imagine.

The reason for this is of course the much better power/performance factor than x86 (at least at the moment).

Power bills for 24/7 server farms are horrendous .. millions of dollars can be saved every year (and the planet) by this sort of migration ..

And I'm not really sure that 8 2.4GHz cores at 45w TDP is anywhere near what you're suggesting as "savior of the planet". The Xeon D 1540, with 8 2.0GHz cores and a 45W TDP will run 16 threads and turbos up to 2.6, and is already "last year's" technology.

If you wanted a low power high performance storage platform, I'd look to the modern Xeon D. Some of the boards that are coming out appear designed for the task, such as the Supermicro X10SDV-7TP4F that I've been trying to get ahold of, which ought to make a killer NAS. There are lower wattage, lower performance parts as well, which would neatly fill in for non-10G-non-SSD filers.
 

Bernard Mentink

Contributor
Joined
Apr 2, 2016
Messages
193
FreeNAS definitely isn't "the best server OS out there", though if you want to say it's the best NAS platform out there, feel free. As a server platform, FreeNAS is big, fat, and eats too many resources.


I was being sarcastic .. hence the poking of the tongue .... (could not find a sarcastic smiley)

Actually, it is not that big and FAT, at least the old 9.2.1 Version ... it runs ZFS very well on my small server setup with only 4G RAM ... workload is streaming multiple Plex HD streams and serving files and web sites.....
Maybe if it has got big and fat with recent versions, the question should be asked ... why? especially if 9.2.1 works so well.

As for an ARM port, it'd be nice to see, but with limited developer resources, it doesn't seem that likely. Further, ZFS specializes in torturing the CPU for things like RAIDZ calculations, so an ARM version might be very much less powerful than you imagine.

I would not call 10,000 VAX DMIPS, "less powerful" ..

As for an ARM port, it'd be nice to see, but with limited developer resources, it doesn't seem that likely. Further, ZFS specializes in torturing the CPU for things like RAIDZ calculations, so an ARM version might be very much less powerful than you imagine.


How hard can it be?, there is already a FreeBSD port on which FreeNAS is based. I would suggest most of the time would be setting and maintaining a separate repo ..


And I'm not really sure that 8 2.4GHz cores at 45w TDP is anywhere near what you're suggesting as "savior of the planet". The Xeon D 1540, with 8 2.0GHz cores and a 45W TDP will run 16 threads and turbos up to 2.6, and is already "last year's" technology.

I did not quote that board as an example, someone else did. The boards I am thinking about are more like 12W TDP and the AMD’s A1100 SoC is 25W TDP, and AMD's TDP is worse case unlike Intel.
http://www.extremetech.com/extreme/...d-processor-the-opteron-a1100-is-finally-here
The A1100 SOC also has a co-processor to offload encryption/decryption and compression/decompression ..
Somewhat useful for ZFS ...

Your Avartar is very appropriate .. :rolleyes:
 
Last edited:

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
I was being sarcastic .. hence the poking of the tongue .... (could not find a sarcastic smiley)

Actually, it is not that big and FAT, at least the old 9.2.1 Version ... it runs ZFS very well on my small server setup with only 4G RAM ...

256MB is a typical server VM around here. I would say 16x is big and FAT.

Also, 9.2.1 specifies 8GB as the minimum amount of RAM, and you're lucky you haven't run into a situation that requires it. Make sure your filer doesn't crash or panic.

workload is streaming multiple Plex HD streams and serving files and web sites.....
Maybe if it has got big and fat with recent versions, the question should be asked ... why? especially if 9.2.1 works so well.

I bumped up the minimum memory requirement for FreeNAS from 6GB to 8GB back in the FreeNAS 8 days, specifically because we're aware of cases where it doesn't work so well. (Do note: I'm probably not the person you want to argue with on memory sizing. You won't get anywhere pleasant.)

I would not call 10,000 VAX DMIPS, "less powerful" ..

No one's seriously measured CPU performance in comparison to a VAX for twenty years.

How hard can it be?, there is already a FreeBSD port on which FreeNAS is based. I would suggest most of the time would be setting and maintaining a separate repo ..

The issue with FreeNAS is that it is very tied in with the hardware and device drivers. ONLY a small set of hardware truly works well in the sort of totally-problem-free way that a NAS or SAN device requires. We know, for example, that there are over a billion problem-free hours attributed to HBA chipsets like the LSI 2008 and Intel gigabit ethernets. Once you start branching out into alternative hardware, you start getting Realtek and Marvell silicon involved in many cases, which we know to be suboptimal.

I did not quote that board as an example, someone else did. The boards I am thinking about are more like 12W TDP

Fine, the Xeon D's go down into the 20's. The Avoton C2750's at 20W. The C2550's at 14W. The C2350's at 6W. And they've been available for like two years.

Your Avartar is very appropriate .. :rolleyes:

Well I'm glad you think so.

Anyways, there's not really anything stopping you from trying to do an ARM port if you want to. I think it's impractical. I like to be proven wrong.
 

Bernard Mentink

Contributor
Joined
Apr 2, 2016
Messages
193
No one's seriously measured CPU performance in comparison to a VAX for twenty years.

Ha, every CPU manufacturer gives a DMIPS for their CPU or SOC even today .... it is useful, because you can then make relative performance comparisons since the beginning of time ... :D
.. and yes, I know there are better benchmarks now ...
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
Ha, every CPU manufacturer gives a DMIPS for their CPU or SOC even today .... it is useful, because you can then make relative performance comparisons since the beginning of time ... :D
.. and yes, I know there are better benchmarks now ...

"every CPU manufacturer gives a DMIPS for their CPU or SOC even today".

Huh. I'm going to call B.S. on that.

Prove me wrong. Show me Intel's DMIPS numbers for:

Xeon E3-1230 v2

Avoton C2750

Xeon D-1540

Bear in mind that I expect these to be numbers provided by Intel. On Intel's website, or some other credible basis for saying "every CPU manufacturer gives."

DMIPS stopped being relevant around 1990. Prove me wrong.
 

Bernard Mentink

Contributor
Joined
Apr 2, 2016
Messages
193
Ok, so you pick on Intel which is the exception these days.

AMD do: http://search.amd.com/en-us/Pages/results-all.aspx#k=Dhrystone
ARM do: http://infocenter.arm.com/help/topic/com.arm.doc.dai0273a/DAI0273A_dhrystone_benchmarking.pdf
http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.faqs/ka3885.html
MIPS cores: https://imgtec.com/mips/classic/
NVIDIA do: http://www.nvidia.com/docs/IO/116757/NVIDIA_Quad_a15_whitepaper_FINALv2.pdf

That's just after a 1 minute google ... there will be lot's more ..
I do realize DMIPS is pretty much irelevent these days as it fits in most CPU's cache, there are much better benchmarks like SPECmark, COREmark, PASSmark etc

Anyway, you are sidetracking the issue by nit-picking. Please remove Intel blinkers.
Show me an Intel or AMD core/soc that even approaches the spec/performance of the 2.5Ghz/48core ThunderX range of ARM64 processors .. prove ME wrong.

The mere fact that that processor range exists is reason enough for this thread and poll. Let's hear what other folk have to say ..
 
Last edited:

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
Ok, so you pick on Intel which is the exception these days.

I'll be sure to note that the widely acknowledged king of the processor world is the exception to "every CPU manufacturer."

Noted.

(crumple, toss, score)

AMD do: http://search.amd.com/en-us/Pages/results-all.aspx#k=Dhrystone
ARM do: http://infocenter.arm.com/help/topic/com.arm.doc.dai0273a/DAI0273A_dhrystone_benchmarking.pdf
http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.faqs/ka3885.html
MIPS cores: https://imgtec.com/mips/classic/
NVIDIA do: http://www.nvidia.com/docs/IO/116757/NVIDIA_Quad_a15_whitepaper_FINALv2.pdf

That's just after a 1 minute google ... there will be lot's more ..
I do realize DMIPS is pretty much irelevent these days as it fits in most CPU's cache, there are much better benchmarks like SPECmark, COREmark, PASSmark etc

Make up your mind. DMIPS stopped being relevant when its results became so distorted it could no longer accurately represent reality. I suppose if you try to find little tiny cores without all the grunt that maybe it would have some value within that subset of CPU's, but basically only because you've stepped back into the dark ages.

Anyway, you are sidetracking the issue by nit-picking. Please remove Intel blinkers.

I have no "Intel blinkers." Anyone who does a little research would know I'm actually an AMD Opteron fan, disappointed by the turn of events in the last ~5-10 years.

Show me an Intel or AMD core/soc that even approaches the spec/performance of the 2.5Ghz/48core ThunderX range of ARM64 processors .. prove ME wrong.

Feel free to send me a board and I'll be happy to test it. Until then, I'll have to believe that your ThunderX cores perform barely better than an Atom D525, and significantly less well than the Avoton parts, because that's what the benchmarks say. The ThunderX is likely to have an advantage in massive parallelism, but that's useless on a typical NAS.

The mere fact that that processor range exists is reason enough for this thread and poll. Let's hear what other folk have to say ..

Nobody's saying all that much because this isn't really all that exciting. Maybe come back when some worthwhile product is actually available on the market. Until then, it's unlikely that we're going to get all excited about this.
 

Bernard Mentink

Contributor
Joined
Apr 2, 2016
Messages
193
Nobody's saying all that much because this isn't really all that exciting. Maybe come back when some worthwhile product is actually available on the market. Until then, it's unlikely that we're going to get all excited about this.

Ok, but at least I am only giving my opinion ... you are trying to speak for everyone ....
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
Ok, but at least I am only giving my opinion ... you are trying to speak for everyone ....

I am? Did I mistakenly lock this thread? (looks) No, thread not locked... no one's forcing everyone else not to speak up. So don't be ridiculous.
 

Bernard Mentink

Contributor
Joined
Apr 2, 2016
Messages
193
Not the point. I was referring to the your following comment, I even highlighted the word for you .... maybe you missed it.
Until then, it's unlikely that we're going to get all excited about this
 

Robert Trevellyan

Pony Wrangler
Joined
May 16, 2014
Messages
3,778
As a former proud owner of an Acorn Archimedes, and correspondingly enthusiastic fan and assembly-level programmer of the ARM from back when those initials stood for "Acorn RISC Machine", I would like to state loudly and clearly that I would prefer that nobody from iX spend any time on an ARM port. Doing so would divert valuable resources away from the core project. Being an open source project, others are, of course, free to develop whatever ports take their fancy.
How hard can it be?
:rolleyes:
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
Not the point. I was referring to the your following comment, I even highlighted the word for you .... maybe you missed it.

I'm not sure what's going on down under that you would think that my use of the word isn't representative of the vast enthusiasm displayed in this thread for your idea. That's in large part because the regular participants here understand that what makes for a good FreeNAS system isn't just a single component - the CPU - but rather a properly blended mix of a well-supported HBA, a well-supported network driver, and other qualities such as actual availability of commodity hardware. I can show you good quality choices for Intel based FreeNAS system boards at a variety of price/performance points, and have in fact offered such examples.

So far what's been suggested for ARM boards is a Gigabyte board that doesn't appear to be available for retail sale, and some "relatively cheap server boards" that also aren't available for sale and for which critical details such as information on ethernet chipset, etc., aren't available.

Regardless, I've already indicated that you're welcome to attempt a port to the hardware.

As with every other response you've made in this thread, I fully expect you to ignore the points I've made and questions I've asked and try some other angle to avoid answering.

So. Since I don't really enjoy debating someone who simply ignores uncomfortable truths and inconvenient facts, ... please stop trolling.
 

Bernard Mentink

Contributor
Joined
Apr 2, 2016
Messages
193
I'm not sure what's going on down under that you would think that my use of the word isn't representative of the vast enthusiasm displayed in this thread for your idea.
This thread has only been up a couple of days, you are being a bit premature .... and ~50% vote for ARM (so far) doesn't exactly show disinterest ...

... please stop trolling.

I started this thread .... and now you are saying I am trolling my own thread? unbelievable!

I, like anyone else, are entitled to have an opinion, just because it isn't the same as yours doesn't make it wrong, or trolling!
Get off your high horse my friend and let the poll run without trying to bias it in your favor.
 

Ericloewe

Server Wrangler
Moderator
Joined
Feb 15, 2014
Messages
20,175
This thread has only been up a couple of days, you are being a bit premature .... and ~50% vote for ARM (so far) doesn't exactly show disinterest ...



I started this thread .... and now you are saying I am trolling my own thread? unbelievable!

I, like anyone else, are entitled to have an opinion, just because it isn't the same as yours doesn't make it wrong, or trolling!
Get off your high horse my friend and let the poll run without trying to bias it in your favor.
  1. Sample size is tiny and uncontrolled. It's statistically meaningless.
  2. Even if statistically valid, any results are essentially irrelevant.
  3. There's a wide gamut between "It would be nice" and "FreeNAS should only exist for ARM64!" that might pick "yes". It's a question that is devoid of any usefulness.
Finally, there are very good technical reasons why ARM is a waste of time for FreeNAS at the moment, as jgreco has eloquently explained. Most importantly, the power efficiency does not scale at all and CPU power is dwarfed by literally every single other major component in a server 90% of the time. Power gains are much easier elsewhere (NICs, HBAs, IPMI/BMCs) with much less pain.

You could say we're being negative, but we're just being realistic. We all want ponies and magic power reductions, but we're not getting them.
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
I started this thread .... and now you are saying I am trolling my own thread? unbelievable!

I, like anyone else, are entitled to have an opinion, just because it isn't the same as yours doesn't make it wrong, or trolling!

You're welcome to your opinion. However, if you enter into a discussion, and then you disregard reasonable counterpoints repeatedly while trying to make new points that are not really relevant, I define that as trolling-like behaviour. As a moderator, I can ask you to kindly refrain from trolling. Trolling has nothing to do with whether it's your own thread or whether or not you're allowed to have an opinion. It's a behaviour.

If I wanted to bias the poll in my favor, I'd just go and edit the damn results. Think on that.
 

Jailer

Not strong, but bad
Joined
Sep 12, 2014
Messages
4,974
freenasarm_zpsbeates1e.jpg
 
Status
Not open for further replies.
Top