Helios4 -- community developed ARM 4 bay NAS device received important upgrade

    • Helios4 -- community developed ARM 4 bay NAS device received important upgrade

      Hi all,

      since some might think this is evil advertising for yet another commercial NAS box... it is not. It's about Helios4, an ARM based 4 bay NAS project currently on Kickstarter: cnx-software.com/2017/05/11/he…id-and-more-crowdfunding/



      Development started last year and the project lead asked in Armbian forum for feedback (which they got and incorporated already). These guys are also fully committed to Open Source goals/community and will even donate to certain OSS projects (OMV amongst it) if the project gets funded.

      But unfortunately Kickstarter didn't allow them to list prices in US dollars and since they're Singapore based and have to show main prices in Singapore dollars majority of people thinks Helios4 would be overpriced. It's not and you have to check the numbers carefully (for example first bundle is listed as '195 S$' with the US dollar price shown one line below in brackets: USD$139. Same applies to shipping costs, there the USD$ value is also just 2/3 of the S$ rate)

      It's only 8 days left but Helios4 came up with a really cool new bundle. This is the first ARM based low cost NAS that allows the use of ECC memory:
      kickstarter.com/projects/helio…open-source/posts/1907929

      ECC memory is important if you care about data integrity since DRAM errors happen every now and then (we have a lot of customer's servers in a monitoring and even servers that survived 72 hours memtester 'burn in' show recoverable 1 bit errors from time to time that get corrected since real servers use at least primitive ECC memory able to recover from single bit errors and reporting also unrecoverable 2 bit errors even if data gets corrupted in such cases).

      Please keep in mind: if your box does not use at least primitive ECC DRAM you won't even take notice of such bit flips if they happen and experience what's called 'silent data corruption' or 'bit rot' (only the severe bit flips that lead to crashes, freezes or file system corruption maybe get some attention). Some further information wrt DRAM bit flips: linux-sunxi.org/Sunxi_devices_as_NAS#cite_note-1

      How does Helios4 relate to OMV? I added few weeks ago an universal OMV installation routine to Armbian's build system. Armbian currently already supports two ARM devices that rely on exactly the same Marvell ARMADA 385 SoC (in fact Helios4 will use exactly those MicroSOM modules from Solid-Run) and I even collected performance numbers already with my Clearfog Pro: forum.armbian.com/index.php?/t…findComment&comment=29574 (in other words: By the time Helios4 gets shipped there will be two perfectly optimized OMV builds for this NAS box, one using Marvell's 4.4LTS kernel and the other latest and greatest mainline kernel with even better performance!)

      So anyone searching for a performant NAS box fully committed to Open Source with up to 4 real and really fast SATA 3.0 ports (there's 2 x USB3 too!) able to operate in a very reliable mode (unlike some USB3 'NAS solutions' pushed out by dev board vendors recently) and also with very low idle consumption if disks are allowed to spin down should have a look. Beware: It's only 8 days left so better check the Kickstarter now and help the project getting funded!
      'OMV problems' with XU4 and Cloudshell 2? Nope, read this first. 'OMV problems' with Cloudshell 1? Nope, just Ohm's law or queue size.

      The post was edited 1 time, last by tkaiser ().

    • 139 USD bucks for 2 GB ECC, Dual Core, high clock, 4x SATA, Gbit ...

      is not cheap but worth it, but for similar price you can get a Intel Celeron J3160 board with external power supply. (Example: ASRock J3160DC-ITX)

      Well without RAM or ECC

      Power consumption might be a little lower with Helios, but 1, 2, 3, 4 Watts don't make big difference on electricity bill.
      Chaos is found in greatest abundance wherever order is being sought.
      It always defeats order, because it is better organized.
      Terry Pratchett

      The post was edited 1 time, last by riff-raff ().

    • riff-raff wrote:

      139 USD bucks for 2 GB ECC, Dual Core, high clock, 4x SATA, Gbit ...
      Please be aware that ECC DRAM is only part of the 'high end' bundle (200 USD$) and that we're talking about a 'whole' DIY NAS solution here (enclosure, power suppy, all necessary cabling included). You need similar bundles to compare with (eg. ODROID-XU4 with Cloudshell 2 featuring similar NAS performance but way less reliable starting at 126 USD$ without shipping, VAT, customs).

      I know that most people don't give a sh*t about energy efficiency but I do and for me a NAS that should run 24/7 idling below 5W with disks in standby/sleep makes a real difference compared to something that wastes 20W while doing nothing (not thinking about electricity bill here but just the less energy wasted the better, another reason I try to avoid RAID at home and why my small NAS boxes all consume less than 1W in idle).

      In my part of the world (Germany) we get HP Microservers for insanely low prices so only basing on price/performance ratio the ECC DRAM equipped Helios4 has a hard time competing with the smaller Microservers but I learned just recently that this is not the case everywhere and you don't get whole 4 bay NAS boxes with ECC memory for less than 300 or 400 USD$. Then Helios4 is a true bargain compared to that if you're really interested in data integrity (most people are not or not even aware that 'bit rotting' happens just while they read these lines ;) )
      'OMV problems' with XU4 and Cloudshell 2? Nope, read this first. 'OMV problems' with Cloudshell 1? Nope, just Ohm's law or queue size.
    • Wolf2000 wrote:

      Der Preis ist aber schon hoch ??
      ('too expensive ??').

      No, I don't think so, the prices are pretty reasonable after you do the necessary currency conversion (still: this is Singapore and not US dollars!).

      The used SoC (Armada 38x -- high speed design with several special acceleration engines) is not really cheap and same also applies to the 'heart' of this NAS solution: Solid-Run's MicroSOM module combined with the NAS baseboard.
      'OMV problems' with XU4 and Cloudshell 2? Nope, read this first. 'OMV problems' with Cloudshell 1? Nope, just Ohm's law or queue size.
    • I am with you, wasting of energy is a thing we have to think about, but it should be in relation to the comfort benefit which the device offers. For example I let my WD Reds run all day long without spindown or parking, that costs energy, but the negative effects of spinup and spindown on the lifetime of a disk and its price compensates that in my oppinion.
      There are other components in a usual household which offer more potential to save energy. (lighting, freezer, washing machine, dishwasher, air conditioning and the frequence of use of these things)


      Müsste ich ein neues NAS aufsetzen würde ich heute auch den hier nehmen: HP ProLiant MicroServer Gen8

      Trotzdem ist der Helios ein hübsches Konzept dem ich den Erfolg durchaus wünsche.
      Chaos is found in greatest abundance wherever order is being sought.
      It always defeats order, because it is better organized.
      Terry Pratchett
    • riff-raff wrote:

      Power consumption might be a little lower with Helios, but 1, 2, 3, 4 Watts don't make big difference on electricity bill
      We're talking about below 5W for the Helios4 vs. +20W for a HP Microserver doing nothing (with normal PCs it's much more worse).
      'OMV problems' with XU4 and Cloudshell 2? Nope, read this first. 'OMV problems' with Cloudshell 1? Nope, just Ohm's law or queue size.
    • ness1602 wrote:

      I'm still for x64 platform,only because of performance.
      Which performance? If you're talking about 'NAS performance' there's not much to gain with 1GbE anyway. Helios4 will saturate a GBit link perfectly just like most x64 designs today.

      BTW: the only device with higher performance for home/SOHO NAS purposes at a reasonable price is also an ARM device based on another Marvell SoC: EspressoBin cnx-software.com/2016/09/23/ma…r-39-and-up-crowdfunding/

      (that's due to this Marvell SoC being connected with 2.5GbE to the Topaz switch so when you set up the switch correctly you can have 3 clients almost saturating their GbE links in parallel. Please also note: SATA, USB3 and PCIe performance of these ARM designs is also outstanding. We measured up to 1.5GB/s with 3 fast SSDs in parallel with Armada 38x)
      'OMV problems' with XU4 and Cloudshell 2? Nope, read this first. 'OMV problems' with Cloudshell 1? Nope, just Ohm's law or queue size.
    • For all of you thinking ARM designs would be dog slow (the 'Raspberry experience'). They're not, Armada 38x and 3700 (that's the one on EspressoBin) is insanely fast while still being amazingly energy efficient.

      I collected some numbers some time ago but all my small SSDs lying around for testing are too slow and therefore bottlenecked the test: forum.armbian.com/index.php?/t…findComment&comment=15265
      'OMV problems' with XU4 and Cloudshell 2? Nope, read this first. 'OMV problems' with Cloudshell 1? Nope, just Ohm's law or queue size.
    • I think the main advantage of an Intel architecture over ARM is versatility. An ARM board is quite enough for a file server, but an Intel board will allow you to run VMs and other apps (docker, media server, video surveillance, home automation, and so on).

      There is no clear-cut better platform. It all depends on your requirements.
    • Nibb31 wrote:

      There is no clear-cut better platform. It all depends on your requirements.
      Yes. And with the Helios4 it's clearly about classical NAS use cases with lowest idle consumption possible and for people who care about 'data integrity' its price is hard to beat (ECC DRAM).

      BTW: We're talking here not about a lame Raspberry Pi or Pi clones. Armada 38x is pretty beefy and stuff like 'docker, media server, video surveillance, home automation, and so on' is possible with this design (but video transcoding or such stuff won't work, the 38x is optimized for different use cases)
      'OMV problems' with XU4 and Cloudshell 2? Nope, read this first. 'OMV problems' with Cloudshell 1? Nope, just Ohm's law or queue size.
    • tkaiser wrote:

      ness1602 wrote:

      I'm still for x64 platform,only because of performance.
      Which performance? If you're talking about 'NAS performance' there's not much to gain with 1GbE anyway. Helios4 will saturate a GBit link perfectly just like most x64 designs today.
      BTW: the only device with higher performance for home/SOHO NAS purposes at a reasonable price is also an ARM device based on another Marvell SoC: EspressoBin cnx-software.com/2016/09/23/ma…r-39-and-up-crowdfunding/

      (that's due to this Marvell SoC being connected with 2.5GbE to the Topaz switch so when you set up the switch correctly you can have 3 clients almost saturating their GbE links in parallel. Please also note: SATA, USB3 and PCIe performance of these ARM designs is also outstanding. We measured up to 1.5GB/s with 3 fast SSDs in parallel with Armada 38x)
      Lets compare it to Amd 5350(or similar X86). Much faster speed in compression/decompression(backup,archiving,etc) atleast. I'm not saying ARM is bad,its just he isnt powerful enough,except for the copy jobs.
    • tkaiser wrote:

      ness1602 wrote:

      I'm still for x64 platform,only because of performance.
      Which performance? If you're talking about 'NAS performance' there's not much to gain with 1GbE anyway. Helios4 will saturate a GBit link perfectly just like most x64 designs today.
      BTW: the only device with higher performance for home/SOHO NAS purposes at a reasonable price is also an ARM device based on another Marvell SoC: EspressoBin cnx-software.com/2016/09/23/ma…r-39-and-up-crowdfunding/

      (that's due to this Marvell SoC being connected with 2.5GbE to the Topaz switch so when you set up the switch correctly you can have 3 clients almost saturating their GbE links in parallel. Please also note: SATA, USB3 and PCIe performance of these ARM designs is also outstanding. We measured up to 1.5GB/s with 3 fast SSDs in parallel with Armada 38x)
      Lets compare it to Amd 5350(or similar X86). Much faster speed in compression/decompression(backup,archiving,etc) atleast. I'm not saying ARM is bad,its just he isnt powerful enough,except for the copy jobs.
    • auRoscoe wrote:

      How would this go doing RAID5 with LUKS encryption on top? Would it still do 1000Mbps read or writes?
      No idea (since I would never again play RAID5 especially at home). But the Marvell Armada 38x has a XOR engine to accelerate RAID5/6 and also CESA (crypto engine) and of course TCP/IP offload engines so not much stuff has to happen on the CPU cores anyway. Again: this is not a Raspberry Toy but a device relying on a professional SoC made for exactly this use case: NAS.

      If LUKS can make use of cryptsetup you get +100 MB/s at 30% CPU utilization: forum.armbian.com/index.php?/t…ted-aes-ni/#comment-21028

      Edit: Above result is with an ancient 4.4 kernel and a Clearfog Base (limited to 1.6Ghz), the Helios4 will be actively cooled and will allow the two ARM cores to clock at up to 1866 MHz so CPU utilization will be even lower with Helios4. Am currently building a new OMV image for Clearfog using latest and greatest mainline kernel 4.11.4 (built with GCC 7.1) and might run some benchmarks later (on a Clearfog Pro though so CPU results with Helios4 will always be 17% better)
      'OMV problems' with XU4 and Cloudshell 2? Nope, read this first. 'OMV problems' with Cloudshell 1? Nope, just Ohm's law or queue size.

      The post was edited 1 time, last by tkaiser ().

    • ness1602 wrote:

      Lets compare it to Amd 5350(or similar X86). Much faster speed in compression/decompression(backup,archiving,etc) atleast
      Well, you already confirmed that we don't get (working) ECC DRAM support with this x86 platform so the most closest x86 competitor to Helios4 is still a HP Microserver (as already mentioned: Since those things are pretty cheap here in Germany I would prefer it for most setups but the Helios4 especially with ECC DRAM is maybe the cheapest option for people really caring about data integrity).

      Anyway: I looked up 7-zip results for AMD 5350 (~4600 MIPS -- can you confirm?) and did two quick test with a freshly built OMV 3.0.79 relying on latest and greatest mainline kernel (4.11.4 ATM):

      Shell-Script

      1. root@clearfogpro:~# 7z b
      2. 7-Zip 9.20 Copyright (c) 1999-2010 Igor Pavlov 2010-11-18
      3. p7zip Version 9.20 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,2 CPUs)
      4. RAM size: 1004 MB, # CPU hardware threads: 2
      5. RAM usage: 425 MB, # Benchmark threads: 2
      6. Dict Compressing | Decompressing
      7. Speed Usage R/U Rating | Speed Usage R/U Rating
      8. KB/s % MIPS MIPS | KB/s % MIPS MIPS
      9. 22: 1105 144 747 1075 | 27402 189 1308 2474
      10. 23: 1302 151 876 1327 | 27507 193 1304 2518
      11. 24: 1267 151 901 1363 | 27644 197 1299 2565
      12. 25: 1266 154 936 1446 | 27177 198 1287 2556
      13. ----------------------------------------------------------------
      14. Avr: 150 865 1302 194 1300 2528
      15. Tot: 172 1082 1915
      Display All

      And another quick test relying on NEON instructions:

      Source Code

      1. root@clearfogpro:~# minerd --benchmark
      2. [2017-06-12 06:43:36] 2 miner threads started, using 'scrypt' algorithm.
      3. [2017-06-12 06:43:36] Binding thread 0 to cpu 0
      4. [2017-06-12 06:43:36] Binding thread 1 to cpu 1
      5. [2017-06-12 06:43:39] thread 0: 4098 hashes, 1.42 khash/s
      6. [2017-06-12 06:43:39] thread 1: 4098 hashes, 1.43 khash/s
      7. [2017-06-12 06:43:39] Total: 2.85 khash/s
      8. [2017-06-12 06:43:40] thread 0: 2850 hashes, 1.57 khash/s
      9. [2017-06-12 06:43:40] thread 1: 2859 hashes, 1.58 khash/s
      10. [2017-06-12 06:43:40] Total: 3.15 khash/s
      11. [2017-06-12 06:43:41] thread 1: 1578 hashes, 1.57 khash/s
      12. [2017-06-12 06:43:41] Total: 3.14 khash/s
      13. [2017-06-12 06:43:41] thread 0: 1575 hashes, 1.55 khash/s
      14. [2017-06-12 06:43:46] thread 0: 7767 hashes, 1.57 khash/s
      15. [2017-06-12 06:43:46] thread 1: 7860 hashes, 1.58 khash/s
      16. [2017-06-12 06:43:46] Total: 3.15 khash/s
      17. [2017-06-12 06:43:51] thread 0: 7866 hashes, 1.57 khash/s
      18. [2017-06-12 06:43:51] thread 1: 7881 hashes, 1.57 khash/s
      19. [2017-06-12 06:43:51] Total: 3.15 khash/s
      20. ^C
      Display All
      This is my Clearfog Pro using the same SoC but limited to 1.6 GHz max. With the Helios4 (1866 MHz cpufreq) we would talk about an overall 7-zip score of 2220 MIPS and almost 3.70 khash/s.

      BTW: My need for compression/decompression speed when doing backups is perfectly satisfied by Armada 38x since I do btrfs snapshots on the device and send them through an VPN to different locations (where other ARM devices with btrfs utilizing zlib-9 compression receive the snapshots and store them -- idle consumption of these boxes less than 1W :)

      Archiving happens locally to an older Core i7 box (with ECC DRAM running OpenSolaris and using ZFS) but I never had a look at the performance since irrelevant anyway (rearchiving/restore are the tasks where performance gets interesting)
      'OMV problems' with XU4 and Cloudshell 2? Nope, read this first. 'OMV problems' with Cloudshell 1? Nope, just Ohm's law or queue size.

      The post was edited 1 time, last by tkaiser ().

    • auRoscoe wrote:

      How would this go doing RAID5 with LUKS encryption on top?
      Small update: Here is 'armbianmonitor -u' output from an ODROID-XU4 (considered a good choice for an ARM based NAS by many for reasons I don't completely understand): sprunge.us/AVZE

      Please look there for the strings 'xor: using function' and 'raid6: using algorithm' and then compare with my Clearfog Pro (still limited to 1.6 GHz unlike Helios4 later): sprunge.us/CVcc

      Armada 38x is only a dual core SoC also with lower clockspeeds than octa core ODROID-XU4 but a) easily outperforms the XU4 and b) does this with special engines so CPU cores are free for other stuff.
      'OMV problems' with XU4 and Cloudshell 2? Nope, read this first. 'OMV problems' with Cloudshell 1? Nope, just Ohm's law or queue size.
    • Users Online 3

      3 Guests