We knew it was coming, and Hardkernel has now launched an updated version of the ODROID-HC1, called ODROID-HC2 based on the same Samsung Exynos 5422 board, but instead supporting 3.5″ hard drives.
The device can now be purchased for $54 plus shipping, but you may also consider adding some accessories like a 12V/2A power supply, and the top cover for the enclosure. [Update: Also listed on Ameridroid now]
ODROID-HC2 specifications:
- SoC – Samsung Exynos 5422 octa-core processor with 4x ARM Cortex-A15 @ 2.0 GHz, 4x ARM Cortex-A7 @ 1.4GHz, and Mali-T628 MP6 GPU supporting OpenGL ES 3.0 / 2.0 / 1.1 and OpenCL 1.1 Full profile
- System Memory – 2GB LPDDR3 RAM PoP @ 750 MHz
- Storage
- UHS-1 micro SD slot up to 128GB
- SATA interface via JMicron JMS578 USB 3.0 to SATA bridge chipset
- Case supports 2.5″ or 3.5″ drives up to 27mm thick
- Network Connectivity – 10/100/1000Mbps Ethernet (via USB 3.0)
- USB – 1x USB 2.0 port
- Debugging – Serial console header
- Misc – Power, status, and SATA LEDs;
- Power Supply
- 12V/2A via 5.5/2.1mm power barrel
- Backup header for RTC battery
- Dimensions – 197 x 115 x 42 mm (aluminum cooling frame)
- Weight – 361 grams
The company provides Ubuntu 16.04 with Linux 4.14, and OpenCL support, but other ditributions are available too including Debian, DietPi, OMV (OpenMediaVault) and others such as Armbian which has released stable Ubuntu and Debian server images for HC1 (but it should work for HC2 too). Source code can be found on Hardkernel’s Github account.
Just like the previous model, ODROID-HC2 is stackable, which explains why it does not sell with a top cover by default, so if you want to buy a single system that works out of the box and with a fully closed enclosure, you should add the optional 12V/2A power supply ($5.90) and a clear or black plastic case ($5.00) to your purchase.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress
Good specs, but then again ARM processors might be the bottleneck when using Twofish or Serpent 256 bits and one gigabit worth of full duplex traffic. RPi Zero (with a USB2 SATA adapter) might be a more interesting choice for those who want to save few bucks.
@Jerry
How are ARM processors limiting today? We’re not stuck with the good old Marvell Kirkwood chips any more in NAS appliances. Annapurna is doing a range of ARM based SoC’s specifically for network appliance/NAS use and they support 10Gbps Ethernet. Although in all fairness I’ve yet to see one that can deliver line speed 10Gbps, but they go well beyond 5Gbps.
I doubt this solution will be slow for its intended use, but I’m not a fan of single drive “NAS” solutions.
R
RPI Zero would be good to save few bucks and lose 90% performance of this home NAS 🙂
For any NAS I would’t go with OS on an SDCARD, they fail very quick compared to eMMC.. For SOHO NAS, I wouldn’t go go below ECC capable NAS…
save few bucks, lose few TB 😉
@Spemer
Do you have a personal experience with NAS on SD card? Just saying because I was using NAS with OS on SD for two years. Last uptime was 8 months.
I have lost 2 sd cards and 2 usb sticks with OS on them.
Now I mount / read-only and /var on ssd, seems to do the trick.
BS. You can buy quality SD cards that last a lot longer than el cheapo eMMC used on entry level boards and why would a NAS appliance constantly write to the rootfs anyway? In OMV for ARM for example we use tmpfs for almost everything so a user would need to ‘tweak’ settings to see an SD card wearing out since with default settings it would need decades.
The ‘secret’ when writing to flash memory is keeping write amplification low and this can be done with some simple tricks (good OS images do implement).
And I prefer those since they prevent people from doing stupid things (eg. playing RAID instead of taking care of data protection and implementing a tested backup strategy — every other day over at OMV forum users present great data losses since they confused availability with data protection and didn’t take care of backup but relied on useless RAID)
After spending an immense amount of time with Ubuntu and the HC1 trying to run Ceph Luminous, I’ve determined that due to the 2GB of RAM, the hosts are unsuitable for use as osd/mon devices. As pure OSD devices you might be able to squeak by, assuming you have your mon/mgr/mds running on other hosts.
I’d be very interested to read about your experience because I was planning just this thing and was waiting for this 3.5′ version.
I would of thought a finned or raised clear plastic top case, would of provided more area for cooling, maybe the cooling difference is not that much?
Side note, here in UK mini USB3 ( claimed not proven ) branned USB drives are popping up in Tesco, Wilko etc. Wonder if they will get useful for storage on these Arm mini computers?
@bkero
Interesting to hear your experience. I have a long term plan to try to test Ceph on ARM64v8, f.eks on RK3328 with 4GB RAM.
Forgive my ignorance but what do you mean by ‘f.eks’?
it might be ‘for example’
Huh? Did you ever have a look at the picture above and noticed that the metal acts as heatsink? There are 3 ‘heat sources’ there: The Exynos SoC (connected with thermal paste to the metal), the 12V/5V DC-DC circuitry (spreads some heat to the PCB) and a disk.
Best solution: fix the disk, then flip the enclosure around so the open part is at the bottom and put the whole thing in a closet not too small. If you want to improve HDD heat dissipation buy 0.5mm thick thermal pads, put them at the disk’s sides and cramp all into the enclosure (at least older disks are designed to dissipate heat to their sides — but I’ve not checked this any more within the last decade or so).
@tkaiser
one of the sd cards which died was kingston 32gb and it died faster than 4gb before with the same OS cloned. I was thinging that wear leveling on bigger card will be better, but I was mistaken, god knows what those manufactures do.
so you say eMMC can die as fast as sd, ughhh, never thought about that, I was looking forward some board with eMMC, thanks for the tip, will look for good eMMC.
Sometimes they don’t know what they do since noname SD card vendors often get counterfeit crap inserted into their supply chain early and the end result are counterfeit SD cards (that’s still the main problem with SD cards, the huge amount of fake flash products flooding the market). See also https://www.bunniestudios.com/blog/?page_id=1022
The term ‘noname SD card vendor’ applies also to Kingston, PNY, Verbatim and all the others except those few that own NAND fabs, have own knowledge, do controllers and produce retail products — that’s only Samsung, SanDisk, Toshiba and Transcend to my knowledge. But again: beware of counterfeit stuff! The more expensive cards the more lucrative for fraudsters to sell a mediocre 2GB card as a fake 128GB ‘genuine Samsung’.
My personal SD card recommendation these days is ‘SanDisk Extreme Plus A1’ if you want both reliability and performance (choose the A1 variant then, not that much more expensive but a lot higher random IO performance which is important with our use cases).
But prior to usage it’s important to test each flash media with either F3 or H2testw for fake flash. Since afterwards the card’s controller thinks all space on the card is used the next step is to use SD Association’s SD Card Formatter. Not to format/partition anything but to do the equivalent of a full TRIM operation telling the card controller that the card is entirely empty. The tool sends DISCARD (CMD38) for the entire capacity to the card’s controller which helps greatly with wear leveling, garbage collection and performance from then on.
And afterwards you need to take care that write amplification remains low (in Armbian we do this with log2ram keeping /var/log in RAM and saving contents back to card only every hour which greatly reduces the real amount of data written and in OMV we take this to the next level with the flashmemory plugin using a mechanism called folder2ram taking care of other areas with high write activity that get stored in RAM only by default)
Funny, just got that A1 Sandisk back from warranty claim. Died after halt a year in dashcam.
@tkaiser
wow, a very good summary for sd use, and good point about the fabs.
thanks, I hope others will find this useful too.
@TLS
ARM can’t encrypt at full network speed. Gigabit ethernet allows encrypting 110 Megabytes / second while decrypting 110 Megabytes per second. Sometimes more if the content is cached and protocol compression is used. 10 Gigabit ethernet means you need to be able to encrypt 1100 megabytes per second. It’s not doable if you don’t have a very high end processor. Serpent and twofish are slower than AES. For example my overclocked Ryzen 1800x can encrypt 390 Megabytes/s. It uses more than 100W TDP and full speed DDR4, not LPDDR3.
@Spemer
You don’t need OS on SD card. RPi can boot OS from USB and after booting you can store it in RAM. A file server is less than 100 Megabytes in size. I have a minimal 20 initramfs image for serving NFS.
forum.armbian.com/topic/4583-rock64/?do=findComment&comment=37829
TL;DR: ARMv8 with crypto extensions can (at least with AES). So get an A53, A72 or better and let the crypto stuff happen inside the dedicated crypto engine. Two known exceptions wrt A53: Raspberry Pi is not capable of this since Broadcom or this mythical foundation forgot to license the crypto stuff and the same applies to Amlogic’s S905 (ODROID-C2, NanoPi K2).
Doesn’t Ryzen 1800x support AES-NI (which will be outperformed by ARMv8/A72 easily at same consumption level)?
@tkaiser
Do you know if the SATA bridge chip used in HC2 does sector emulation or not for 512e HDD? Most USB enclosure out there does this to get 3TB+ HDD working on older OS that doesn’t support them.
By sector emulation I mean this: https://plugable.com/2013/03/21/understanding-large-sata-drive-compatibility/
@tkaiser
Oh, my NAS is RAID-0 for this simple reason.
@Jerry
Says who? What SoC? Annapurna can do it at Gigabit+ speed for sure. Example https://www.qnap.com/en/product/ts-431p2
That’s faster than Gigabit speed.
Realtek, not quite at Gigabit speed, but close enough for most consumers https://www.qnap.com/en/product/ts-228a
Maybe read up on some more recent hardware?
A lot of SoC’s today has a separate crypto co-processor that handles all that shit. Also, how many consumers encrypt their NAS do you think? This product and similar ARM based products are not intended for business use.
Curious as to why they left off the 8-bit eMMC interface on both the HC1 and HC2 boards.
The XU4, HC1, HC2, C1, and C2 all have hardware crypto accelerators. Encrypt/decrypt are memory block operation. ARM designs leverage this and place the crypto hardware at the memory interface level. This is a more optimal design than CPU based crypto that stalls the CPU. There is a slide somewhere on the internet introducing ARM AES extensions where its explicitly stated it is not meant to replace hardware crypto.
As far as I know, Android is the only OS that leverages this hardware. I have not personally used this feature (or Android) and so can not comment on performance.
@crashoverride
Linaro have a blog entry
Accelerated AES for the ARM64 Linux kernel
If it at all helps
If I am not mistaken, kernel 4.14 on the XU4 line introduces AES drivers. But I am not aware of anyone actually testing/using them.
Wishing they offered this with an mSATA option, and a much smaller case…
IMHO they should make a board supporting 5..12V power from the jack, and replace both HC1 and HC2 with this new board. It will make a direct upgrade for both HC1 and HC2 users, allowing them to maintain a single board with just two different heatsinks. Or if their HC1 users accept to migrate to 12V, having just 12V would be fine as well.
Why should we today care about crappy USB-to-SATA bridge chipsets manufactured a decade ago to support Windows versions already outdated back then? 🙂
A SATA disk behind a JMS578 with most recent firmware (check ODROID wiki) behaves exactly the same as on a real SATA port (except performance of course and there’s one minor exception: http://forum.odroid.com/viewtopic.php?f=97&t=29069)
Why? This is USB3-SATA here which doesn’t affect the NAS use case much (since sequential HDD transfer speeds do not decrease noticable) but has some serious drawbacks with SSDs and why those are used (high random IO performance which gets pretty much destroyed by the added USB layer). I think I already suggested it to you since you ask for the same thing now for over a year: with the two more expensive i.MX6 SoMs HummingBoard Pro supports mSATA, HummingBoard Edge provides M.2 (but i.MX6 random IO storage performance isn’t that great either — as usual all numbers in Armbian forum).
@tkaiser, I’m curious, are you doing the armbian + OMV releases, or do you do the OMV releases and happen to support armbian.
I’m running on friendlyarm with good success with omv, and need to update that, and I am pretty sure you are doing the one I’m using there.
much appreciated your comments and all the comments by everyone here.
thanks
Contributing to both projects (focused on NAS/server tasks with ARM devices and there focusing on all the necessary performance tweaks mostly since it’s so sad that the majority of OS images doesn’t take care of appropriate settings which is the reason many devices perform so poorly).
Armbian and those OMV images for ARM devices (the one for Raspberry Pi included in the meantime) are all the result of a 100% fully automated build everyone on this planet is able to repeat. Though all ‘official’ OMV images for ARM devices have been built on my MacBook while all official Armbian images were built on Igor’s PC. But this doesn’t matter since all of this is 100% reproducible (every single performance tweaks gets part of the build system). Details: forum.armbian.com/topic/3953-preview-generate-omv-images-for-sbc-with-armbian/?do=findComment&comment=29577
In case you currently use FriendlyARM’s OMV builds… they were created differently but AFAIK also contain almost all of the performance tweaks outlined in the thread referenced above.
Qnap have a Realtek RTD1295 quad-core processor Nas, the TS-128A , 1GB ram, but it is £130 GBP @ Snap for just 1 bay! Makes above seem cheap.
I bought one not too long and used OMV 3.x third party images. Gets about 960 Megabits / sec with iperf, and with ext4 writes to drive start out at about 109 MegaBytes/Sec and drop down to about 50/20 MegaBytes after 10 minutes (I was testing a 100G file). This is with spinning rust. Overall its quite nice, and when you run ALL cores at max with sysbench, they get up to 85c, but then nicely settle down to 41c when running ‘idle’. Its really quite an impressive board. Only annoying thing about armbian is they don’t enable auto dns registration, so you have to add dhcpcd and modify network manager config. Overall is a great board.
At least with the new N1 board it seems they listened to you (most probably not since they started the design back in May last year): ‘If you don’t use 3.5″ HDD, the input voltage can be very wide like 6 ~ 18Volt. Once you connect any 3.5″ HDD, you have to use 12volt.’
No more annoying undervoltage hassles at least with N1 🙂
Well, if this is with a 3.5″ HDD then there’s something wrong since even on their slowest media zone those disks should be good for +60 MB/s if it’s a slow one. And the SATA bridge with appropriate settings is able to push 380 MB/s constantly (though I test with 12GB maximum usually and not 100)