Stack-able micro computers

Moderators: meveric, mdrjr, odroid

Re: Stack-able micro computers

Unread postby Raymond Day » Sat Aug 12, 2017 11:36 am

Wow very good. That's all I been using mine for is a server. Nice to have SATA right on it.

I use UCtech USB 3.0 to SATA Converter if can find that any more. But it works real good with the USB cable right in it and if you take the hard drive out and put it on some other SATA you can still read it. Some USB to SATA format it some way were you put it in some other SATA can't read it has to reformat it.

So this ODROID-HC1 is something I been looking for a long time. I don't need any HDMI to use it as a headless server.

WDLabs has a PiDrive No SATA ports the board right on it is a USB.

When this ODROID-HC1 comes out I like to order one.

Thank you for making it.

-Raymond Day
Raymond Day
 
Posts: 376
Joined: Tue Jun 04, 2013 8:18 pm
Location: Belleville MI
languages_spoken: english
ODROIDs: Odroid-U2, XU4, C1+,

Re: Stack-able micro computers

Unread postby SvenDowideit » Sat Aug 12, 2017 3:26 pm

Very timely!

I work on RancherOS - http://rancher.com/rancher-os

and while we biuld rpi and rpi64 releases, it takes a long time to build the 32bit version - a small cluster of odroid would not only speed up the builds, but I'd port RancherOS to it too :D
SvenDowideit
 
Posts: 1
Joined: Sat Aug 12, 2017 1:56 pm
Location: Brisbane, Australia
languages_spoken: english
ODROIDs: c1, xu4

Re: Stack-able micro computers

Unread postby synportack24 » Sun Aug 13, 2017 12:41 am

The HC seems like an awesome use of the XU3/4 platform. I'm really excited to see what images come out of the community to support it! Are there any official OS planned from Hardkernel apart from the typical ones for the XU4? Can't wait till August 21st
http://magazine.odroid.com/
submit an article get cool stuff!
User avatar
synportack24
 
Posts: 105
Joined: Sat Jan 10, 2015 10:16 am
Location: 대한민국 - 안양시
languages_spoken: English, German, Korean
ODROIDs: C0, C1, C1+, C2, XU4

Re: Stack-able micro computers

Unread postby MrDreamBot » Sun Aug 13, 2017 7:19 am

On a different note, I suggest the My Cluster should have an option to include the My Cloud at the bottom of the stack and 3 My Cluster boards/cases on top (do they fit together?). This is because, from the photo, the My Cluster boards don't have either USB3 connector nor SATA making it tricky to include a fast storage for the cluster which is mandaory, in my opinion for a cluster. By providing this option, customers can choose either the My Cluster as is or the one with My Cloud (SATA storage) if they don't already have a a file server. Just my 2cents.
User avatar
MrDreamBot
 
Posts: 24
Joined: Fri Mar 13, 2015 10:58 am
languages_spoken: english
ODROIDs: ODROID-C2, ODROID-C1/C1+, ODROID-XU3, ODROID-XU4, ODROID-MC1, ODROID-HC1,ODROID-U3

Re: Stack-able micro computers

Unread postby MrDreamBot » Mon Aug 14, 2017 6:13 am

One question: do we really need the fan at the end of the My Cluster stack? Isn't the metal case enough to dissipate the heat? I really like fanless (meaning quiet) computers and clusters. The only thing I don't like about the XU4 is the noisy fan!
User avatar
MrDreamBot
 
Posts: 24
Joined: Fri Mar 13, 2015 10:58 am
languages_spoken: english
ODROIDs: ODROID-C2, ODROID-C1/C1+, ODROID-XU3, ODROID-XU4, ODROID-MC1, ODROID-HC1,ODROID-U3

Re: Stack-able micro computers

Unread postby mad_ady » Mon Aug 14, 2017 3:13 pm

For high loads and performance you'll need the fan or the cores will throttle. If you want it passive you won't plug the fan. But since it's large and runs at constant speed, it will probably not be as annoying as the stock fan.
User avatar
mad_ady
 
Posts: 2400
Joined: Wed Jul 15, 2015 5:00 pm
Location: Bucharest, Romania
languages_spoken: english
ODROIDs: XU4, C1+, C2

Re: Stack-able micro computers

Unread postby gabm » Mon Aug 14, 2017 3:21 pm

Great Idea! We've wished to have such systems available at times. We're working on software for automotive applications and while our software can handle rapid prototyping demands very well, finding suitable and rapidly-deployable hardware has always been cumbersome. A key selling point of our software stack is scalability as deployment is easy and rapid. The MC1 seems like a great device for rapid hardware deployment here. It already has casing, uses USB power supply and is supposely low on power usage.. We'd love to get a hand on some MC1 devices and our share feedback with you.
gabm
 
Posts: 1
Joined: Mon Aug 14, 2017 3:03 pm
languages_spoken: english
ODROIDs: XU4

Re: Stack-able micro computers

Unread postby Armand79th » Tue Aug 15, 2017 11:46 pm

So.... how do I get involved in the testing process? :)

I'm currently employed with a large webhosting company and I'd love to test some of these boards as part of a smaller webhosting or application cluster.
Armand79th
 
Posts: 1
Joined: Sat Aug 12, 2017 8:53 pm
languages_spoken: english
ODROIDs: U3

Re: Stack-able micro computers

Unread postby ksubox » Wed Aug 16, 2017 10:39 am

Sounds interesting, but I can't find answer how to make real home computer system.

Actually what we need:
1. Network storage - not huge, but quite big for family/work history - and I can't loose it, so must be at least RAID1
2. Servers for family & work - web/mail/ftp..., git repositories - probably also at least RAID1 & memory?
3. Application servers for my public services - probably cluster solution is good for that if interconnection is fast, but I wonder how good could be this solution for Java application servers ? 2M memory is quite small for Java VM + application data, so we will have to distribute data across cluster and as a result - need high speed interconnection/processing.

If somebody could explain how to design what we need, it would be very helpful.

Thank you.
ksubox
 
Posts: 29
Joined: Wed Jan 15, 2014 12:42 pm
languages_spoken: english
ODROIDs: ODROID-X2, X4, C2

Re: Stack-able micro computers

Unread postby KRYPTALIVIAN » Thu Aug 17, 2017 12:09 am

When someone gets their hands on a unit, can you measure the wattage when under a load(with HDD attached) please :)
KRYPTALIVIAN
 
Posts: 1
Joined: Thu Aug 17, 2017 12:06 am
languages_spoken: english
ODROIDs: C1 XU4

Re: Stack-able micro computers

Unread postby odroid » Thu Aug 17, 2017 4:08 am

@ksubox,
If you really need a RAID1 storage, you can consider our CloudShell2.
Or two HC1 with scheduled RSYNC can be a perfect backup plan.
A lot of Light-weight home server applications can run with 2GB smoothly.
But if you need a high performance server, you have to consider a much more expensive server system instead of $49.

@KRYPTALIVIAN
Heavy HDD access (100MB/sec) + Heavy Gbit access (100MB/sec) + Full 8-core CPU stress test = 14~16Watt (5V/3A)
In Idle mode, it could be down to 3.5~5Watt.
User avatar
odroid
Site Admin
 
Posts: 24367
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English
ODROIDs: ODROID

Re: Stack-able micro computers

Unread postby elatllat » Thu Aug 17, 2017 4:15 am

With sata instead of USB 3 fewer drives can be used so I'll avoid the HC1.
The XU4 is happy with at least 23 drives.

There are very few practical reasons to use a cluster.
For personal use some people will likely get them just for playing with, but proper fail-over uses nodes in different data centers.
Data-centers like scaleway tend to make there own stuff instead of stacking others, but it would be nice to see more arm cloud offerings.

UAS and S.M.A.R.T. compatibility are drive specific and UAS tends to only need blacklisting for the larger drives (8TB Segate)

@ksubox The instructions are about the same for any Linux, but it's all off topic so make a new thread for each question.
elatllat
 
Posts: 561
Joined: Tue Sep 01, 2015 8:54 am
languages_spoken: english
ODROIDs: XU4

Re: Stack-able micro computers

Unread postby tkaiser » Thu Aug 17, 2017 3:52 pm

KRYPTALIVIAN wrote:When someone gets their hands on a unit, can you measure the wattage when under a load(with HDD attached) please :)


Measuring behind ODROID's 4A PSU I got up to ~11W with typical NAS benchmarks but using an SSD instead of a 2.5" HDD (so add a little bit more especially with random IO loads, if you do sequential IO with a somewhat decent HDD -- streaming a movie for example -- consumption is very low). This was with cpufreq/dvfs defaults (allowing to clock big/little CPUs with 2.0/1.4GHz) and Hardkernel's dvfs OPP mean a heavily increased consumption when climbing up to highest clockspeeds especially on the big cluster.

Since based on previous and yesterday's monitoring with appropriate IRQ/CPU affinity settings seeing nowhere a CPU bottleneck my personal goal currently is to develop settings that work as fast as defaults but cutting cpufreq on the big cores to 1.4Ghz as well. First tests show 3W less then with just 5% performance drop when using synthetical benchmarks (most probably zero difference when copying files with Windows Explorer or macOS Finder due to some magic happening in the background explained here http://www.helios.de/web/EN/support/TI/157.html )

I shared some first (excellent) experiences at CNX: http://www.cnx-software.com/2017/08/10/ ... ent-545244

For whatever reasons yesterday I got low NAS write performance and only ~550-600 MBits/sec max in RX direction when testing with iperf3 but I'm pretty sure that this must be a problem on my side since RTL8153 on XU4/HC1 is known to reach 940 Mbits/sec. Anyway: If someone running XU4/HC1 with latest 4.9 kernel can provide iperf3 numbers in both directions ('iperf3 -s' on another machine and then 'iperf3 -c' and 'iperf3 -c -R' on the ODROID) that would help.
tkaiser
 
Posts: 95
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: Stack-able micro computers

Unread postby willfe » Fri Aug 18, 2017 7:31 am

I'd love to put a few of the HC1's to the test running a Ceph cluster! I suspect they'd make very good storage nodes, and could possibly also handle running the monitor and/or manager daemons too. For testing I'd use three of them (with a 1 or 2TB disk each) as storage nodes, a fourth one for the monitor & manager, and a fifth for a metadata server (for Ceph's POSIX filesystem support). The monitor & manager are pretty lightweight, so they can probably run together on a node. The metadata server is CPU-intensive though, so it should run on its own node. The OSDs (storage nodes) have moderate CPU requirements but the CPUs on the HC1 should be more than enough.

With gigabit ethernet the resulting storage cluster probably wouldn't be the fastest thing in the world but I bet it could easily keep up with a Drobo. The other benefits of Ceph are just a nice bonus :)

@odroid, are you also sampling the HC1's? I'd be happy to build a Ceph storage cluster on them, document the entire process and provide benchmarks and feedback.
willfe
 
Posts: 4
Joined: Fri Dec 12, 2014 8:23 pm
Location: Orlando, Florida, USA
languages_spoken: english
ODROIDs: U3, C2

Re: Stack-able micro computers

Unread postby uDude » Fri Aug 18, 2017 2:58 pm

With the HC1 can we use SMART tools on the drive?

Also, can we use Samsung ssd built in crypto?
uDude
 
Posts: 59
Joined: Mon Sep 15, 2014 7:32 am
languages_spoken: english
ODROIDs: 3 U3's,
2 XU3's,
5 C1's,
26 XU4's,
3 HC-1s

Re: Stack-able micro computers

Unread postby mad_ady » Fri Aug 18, 2017 3:15 pm

Smart yes. I don't know how samsung encryption works
User avatar
mad_ady
 
Posts: 2400
Joined: Wed Jul 15, 2015 5:00 pm
Location: Bucharest, Romania
languages_spoken: english
ODROIDs: XU4, C1+, C2

Re: Stack-able micro computers

Unread postby tkaiser » Fri Aug 18, 2017 4:31 pm

uDude wrote:With the HC1 can we use SMART tools on the drive?


The JMS578 USB SATA bridge on HC1 fully supports SMART, SAT (SCSI / ATA Translation -- that's why you can also use tools like hdparm without issues) and also TRIM (though no idea about software support. The last time I checked I had the impression discard/TRIM is not available through USB with Linux kernel)

Edit: I forgot to mention that JMS578 will be fully supported by smartmontools starting with version 6.6: https://www.smartmontools.org/ticket/841 (so you need to specify '-d sat' for now manually to access devices behind HC1's USB-to-SATA bridge)
Last edited by tkaiser on Fri Aug 18, 2017 6:19 pm, edited 1 time in total.
tkaiser
 
Posts: 95
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: Stack-able micro computers

Unread postby nobe » Fri Aug 18, 2017 5:29 pm

uDude wrote:can we use Samsung ssd built in crypto?

i suppose you talk about OPAL-compliant Self Encrypting Drives technology.
https://wiki.archlinux.org/index.php/Se ... ing_Drives

as far as i know, the software isn't mature yet in general (i don't find any sedutil package in my x86 ubuntu distro).
for arm linux, it doesn't even seem to compile for now : https://github.com/Drive-Trust-Alliance ... issues/148
nobe
 
Posts: 96
Joined: Sun Feb 07, 2016 9:52 pm
languages_spoken: english, french
ODROIDs: Odroid-C2

Re: Stack-able micro computers

Unread postby fvolk » Fri Aug 18, 2017 11:37 pm

I'm am bit cautious, the 5TB 2.5" Seagate draws a bit more power than smaller 2.5" drives.
I was unable to get it stable with my C2, it appeared to work at first but sometimes it made clicking noises or disconnected - on a PC it works fine.

So I hope you implemented robust power delivery with the HC1, then you have a customer -> I will upgrade my current C2+2TB Seagate that are now running for ~320 days without problems to a HC1 with a 5TB :-)
fvolk
 
Posts: 94
Joined: Sun Jun 05, 2016 11:04 pm
languages_spoken: english
ODROIDs: C2

Re: Stack-able micro computers

Unread postby Seb » Sat Aug 19, 2017 12:13 am

Oh, my! I can't wait to have a working version of my OS to cluster on this... It's a fairly distant dream right now. Perhaps there'll be an Odroid-HC2/MC2/XU5/whatnot by then... one step at a time :)
Seb
 
Posts: 8
Joined: Sun Jul 16, 2017 3:36 am
languages_spoken: english

Re: Stack-able micro computers

Unread postby tkaiser » Sat Aug 19, 2017 12:14 am

fvolk wrote:I'm am bit cautious, the 5TB 2.5" Seagate draws a bit more power than smaller 2.5" drives.


Not really, they work fine on USB3 ports providing 900mA.

HC1 is Hardkernel's answer to the most common disk access problems that occured with XU4 in the past (under-voltage/underpowering, cable/contact issues, crappy USB-to-SATA bridges). On HC1 PCB the 5V go from the barrel jack directly to pins 7-9 on the SATA power connector. The PCB is prepared to feed 3.5" HDDs too (there's a 12V input feeding pins 13-15 on the SATA power connector) so there's no need to worry about HDD powering problems ever again.
tkaiser
 
Posts: 95
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: Stack-able micro computers

Unread postby fvolk » Sat Aug 19, 2017 1:38 am

tkaiser wrote:
fvolk wrote:I'm am bit cautious, the 5TB 2.5" Seagate draws a bit more power than smaller 2.5" drives.


Not really, they work fine on USB3 ports providing 900mA.


The 15mm 3-5TB draw up to 1.2A (http://www.seagate.com/www-content/prod ... 1609gb.pdf) that's more than 0.9A per USB spec. Plus additional load of USB3<-->SATA converter.

In my experience it's been gamble, now I only use them directly attached to a PC.
fvolk
 
Posts: 94
Joined: Sun Jun 05, 2016 11:04 pm
languages_spoken: english
ODROIDs: C2

Re: Stack-able micro computers

Unread postby odroid » Sat Aug 19, 2017 5:38 am

tkaiser wrote:For whatever reasons yesterday I got low NAS write performance and only ~550-600 MBits/sec max in RX direction when testing with iperf3 but I'm pretty sure that this must be a problem on my side since RTL8153 on XU4/HC1 is known to reach 940 Mbits/sec. Anyway: If someone running XU4/HC1 with latest 4.9 kernel can provide iperf3 numbers in both directions ('iperf3 -s' on another machine and then 'iperf3 -c' and 'iperf3 -c -R' on the ODROID) that would help.


We will run the iperf test early next week and let you know the result.
But I believe it should be near 900Mbps since the yesterday's LanTest result was fine with HC1 + HGST_1TB_HDD.
Image
Image

Update: One user reported the network performance is around 940Mbps.
viewtopic.php?f=96&t=27982
User avatar
odroid
Site Admin
 
Posts: 24367
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English
ODROIDs: ODROID

Re: Stack-able micro computers

Unread postby Umbreon » Sat Aug 19, 2017 2:02 pm

@Odroid

Have you got any issue with the hard drive like noise or disconnection or other bad things ?
User avatar
Umbreon
 
Posts: 31
Joined: Sun Nov 22, 2015 2:32 am
languages_spoken: english french
ODROIDs: W, C1, XU4

Re: Stack-able micro computers

Unread postby tkaiser » Sat Aug 19, 2017 9:56 pm

fvolk wrote:
tkaiser wrote:
fvolk wrote:I'm am bit cautious, the 5TB 2.5" Seagate draws a bit more power than smaller 2.5" drives.


Not really, they work fine on USB3 ports providing 900mA.


The 15mm 3-5TB draw up to 1.2A (http://www.seagate.com/www-content/prod ... 1609gb.pdf) that's more than 0.9A per USB spec. Plus additional load of USB3<-->SATA converter.


Hmm... we have reports from 2 customers who bought some of them that they just work fine on USB3 ports (maybe Seagate's enclosure contains capacitors to compensate for the short peak consumption when platters start spinning?). But thanks for the reminder, I forgot to ask whether they've tested with average PCs or Macs (based on some tests I found Apple USB ports providing slightly more than 900mA -- though testing methodology not that scientific ;) ).

Anyway I don't believe there's anything to worry about powering since as already said HC1 is Hardkernel's answer to HDD powering problems of the past. The distance between DC-IN jack and SATA power connector are a few cm and I would be really surprised if Hardkernel used PCB traces too thin or put an underdimensioned current limiter in between.

I thought already about testing with rather power hungry WD Velociraptors (10k rpm) I have lying around... but then remembered that they need power on the 12V rail unlike nearly all other 2.5" disks...
tkaiser
 
Posts: 95
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: Stack-able micro computers

Unread postby ard » Mon Aug 21, 2017 6:36 pm

willfe wrote:With gigabit ethernet the resulting storage cluster probably wouldn't be the fastest thing in the world but I bet it could easily keep up with a Drobo. The other benefits of Ceph are just a nice bonus :)

Why wouldn't it be fast? Each node increases the total performance, unless you swamp the cluster with too many clients.
But san's die too with too many clients. Actually, that happens very fast with san, as they need dedicated very expensive switches, unless you do san over ethernet, like FCoE.
personally I am very happy with a > 60MB streaming performance using FCoE. The hard disks aren't that fast after all.
ard
 
Posts: 55
Joined: Tue Jul 09, 2013 2:12 am
languages_spoken: english, dutch, german
ODROIDs: ODROID-U2

Re: Stack-able micro computers

Unread postby venkatbo » Mon Aug 21, 2017 7:45 pm

A couple of HC1 related observations:

In the HC1 wiki, for the various tables/charts, would it make sense to flip the bars/columns to list the info for HC1 first ... and RPi last. After all you want to highlight your (better) HC1.

In the legends indicate that "longer" bars imply better performance.
User avatar
venkatbo
 
Posts: 444
Joined: Mon Feb 25, 2013 3:10 pm
Location: Bay Area, California
languages_spoken: english
ODROIDs: C0/C1+/C2 U3+ XU3-Lite/XU4 VU/VU7+ Touchscreen Show TFT Cloudshell SmartPower mAHRS DAC GPS Multiscope

Re: Stack-able micro computers

Unread postby willfe » Tue Aug 22, 2017 1:31 am

ard wrote:Why wouldn't it be fast? Each node increases the total performance, unless you swamp the cluster with too many clients.
But san's die too with too many clients. Actually, that happens very fast with san, as they need dedicated very expensive switches, unless you do san over ethernet, like FCoE.
personally I am very happy with a > 60MB streaming performance using FCoE. The hard disks aren't that fast after all.

I'm not saying it'd be terribly slow. You'll just only ever get about 110MB/sec in or out of the system. Inter-node communications also happen at that speed (so spreading data across nodes shouldn't have much performance impact on clients), but actually getting data into and out of the cluster involves using a RADOS gateway (for the object store) or a regular ol' host that mounts a POSIX CephFS from the cluster and exports it via NFS, CIFS, etc. You can set up multiple hosts to do this, of course, but network speed is still going to be the bottleneck, not Ceph or the individual nodes.

As you say, without dedicated network hardware (and a well-designed network) the biggest potential slowdown comes from multiple clients all transacting at once.

It'd still be an awesome setup to try, though :)
willfe
 
Posts: 4
Joined: Fri Dec 12, 2014 8:23 pm
Location: Orlando, Florida, USA
languages_spoken: english
ODROIDs: U3, C2

Re: Stack-able micro computers

Unread postby tkaiser » Tue Aug 22, 2017 2:25 am

Today I looked a little bit closer at throttling performance/behaviour with HC1: https://www.cnx-software.com/2017/08/21 ... ent-545362

TL;DR: throttling 'performance'/strategy with 4.9 is way better than with 3.10 before and I'm wondering whether we could/should increase thermal trip points to let throttling jump in later. Maybe even only when HC1 is detected (missing USB hub) since here's no annoying fan to take care of.

I'm about to push out a new OMV release soon anyway (since with Armbian's kernel we chose 2MB coherent memory pool [1] now and the current OMV image sets just 1MB through kernel cmdline) and want to evaluate changes to thermal trip points before. If they're adjustable from sysfs that would be great since then I could add a simple daemon to OMV monitoring temperatures in the background and able to differentiate between XU4, XU4Q and HC1 and/or using a small config file so experienced users can simply choose at which temperature they want throttling to start. Any ideas/objections?

@odroid: Currently OMV installation routine adds 'ppa:kyle1117/ppa' to deal with your Cloudshell 2 support scripts. You said back then the PPA would move to a more official location. If that's the case or you still plan to please drop me a note so I can a) switch the source and b) think about a way to switch already existent locations to the new PPA/repo. Thanks!

[1] https://forum.armbian.com/index.php?/to ... mory-size/
tkaiser
 
Posts: 95
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: Stack-able micro computers

Unread postby odroid » Tue Aug 22, 2017 3:20 am

@tkaiser,
We want to keep using kyle1117's PPA since there is no big concerns.
He is our key engineer who has worked for Hardkernel over 4 years.

BTW, the HDD temperature could be up to 55C due to high CPU temperature when we tested 3TB transferring on HDD and Ethernet simultaneously for several hours.
viewtopic.php?f=146&t=26016&start=250#p198451
AFAIK, the HDD temperature should be lower than 60C to avoid any data corruption.
Once we release an official plastic cover for HC1 in the end of September, the HDD temperature can go higher.
So I don't like to change the thermal management trip points aggressively.
User avatar
odroid
Site Admin
 
Posts: 24367
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English
ODROIDs: ODROID

Re: Stack-able micro computers

Unread postby odroid » Tue Aug 22, 2017 3:23 am

We could assume the total cost of electricity bill using ODROID-HC1 : Home Cloud One Server 24hours/7days at home.

We installed the Ubuntu 16.04 minimal image(Kernel 4.9.44) and FTP, SAMBA, NFS, Plex-Media, Seafile, Serviio, Transmission and WebDAV server applications were running in background.
It consumes average 3.6Watt when it is idle and the attached hard disk drive is in spin-down mode.

Image

Here is the detail analysis of power consumption.
A peak pulse up to 5V/3A(~15Watt) when the HDD starts to run in the OS booting process.
Idle, HDD is still spinning : Around 5-6Watt.
Transferring some big files with SAMBA transfers via Gbit Ethernet interface : 7~9Watt approximately.
Idle, HDD spin-down : Average 3.6Watt.
If we change the max scaling frequency, we can lower it down to below than 3.0Watt.
But, the system performance will be slightly slower.

Anyway, I believe ~4 watt idle mode power consumption must be acceptable for most users.

We've measured the power consumption of the ODROID-HC1 with our SmartPower2.
User avatar
odroid
Site Admin
 
Posts: 24367
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English
ODROIDs: ODROID

Re: Stack-able micro computers

Unread postby tkaiser » Tue Aug 22, 2017 6:12 am

odroid wrote:@tkaiser,
We want to keep using kyle1117's PPA since there is no big concerns.


Perfectly fine with that. Just wanted to ask since now would be the perfect time to switch if necessary. Wrt your HDD temperature concerns I partially agree (my own tests simulating a tiny enclosure weren't that bad and I did not find that much correlation between SoC temperature difference of 10°C and drive temperature). Let's postpone thermal trip point changes until you made your own tests with the plastic shield. :)
tkaiser
 
Posts: 95
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: Stack-able micro computers

Unread postby joshi » Tue Aug 22, 2017 6:27 am

Have anybody tried this as a streaming device? I don't think it's advisable to use RTSP or RTMP on such setup, but HLS or DASH might be quite nice given you have a decent SSD.
The question is - will it handle streaming load:
1) From disk IO perspective - ~8 mbps write, at least 400 mbps read
2) From network connection perspective - 1 incoming UDP multicast, at least 50 outgoing HTTP

I guess only experiment might show.

On the other hand, in more balanced, home streaming device setup, it might work better, as the needs are different:
1) From disk IO perspective - ~70 mbps write, ~32 mbps read
2) From network connection perspective - either 1 incoming UDP multicast if demux is done locally or up to 20 of them if done externally and 4-5 outgoing HTTP ones.
joshi
 
Posts: 3
Joined: Tue Aug 22, 2017 6:06 am
languages_spoken: english, russian, georgian

Re: Stack-able micro computers

Unread postby Brian.K » Tue Aug 22, 2017 11:15 am

iperf performance on ODROID-HC1
* Network Topology: ODROID-HC1(192.168.100.30) <---> Gigabit Switching Hub <---> HOST PC(192.168.100.2)
Code: Select all
odroid@odroid:~$ uname -a
Linux odroid 4.9.44-54 #1 SMP PREEMPT Sun Aug 20 20:24:08 UTC 2017 armv7l armv7l armv7l GNU/Linux
odroid@odroid:~$ iperf -c 192.168.100.2 -P 10
------------------------------------------------------------
Client connecting to 192.168.100.2, TCP port 5001
TCP window size: 43.8 KByte (default)
------------------------------------------------------------
[ 12] local 192.168.100.30 port 60508 connected with 192.168.100.2 port 5001
[  3] local 192.168.100.30 port 60490 connected with 192.168.100.2 port 5001
[  4] local 192.168.100.30 port 60492 connected with 192.168.100.2 port 5001
[  5] local 192.168.100.30 port 60494 connected with 192.168.100.2 port 5001
[  6] local 192.168.100.30 port 60496 connected with 192.168.100.2 port 5001
[  7] local 192.168.100.30 port 60498 connected with 192.168.100.2 port 5001
[  9] local 192.168.100.30 port 60500 connected with 192.168.100.2 port 5001
[ 10] local 192.168.100.30 port 60504 connected with 192.168.100.2 port 5001
[  8] local 192.168.100.30 port 60502 connected with 192.168.100.2 port 5001
[ 11] local 192.168.100.30 port 60506 connected with 192.168.100.2 port 5001
[ ID] Interval       Transfer     Bandwidth
[  3]  0.0-10.0 sec   125 MBytes   105 Mbits/sec
[  4]  0.0-10.0 sec   125 MBytes   105 Mbits/sec
[  5]  0.0-10.0 sec   125 MBytes   105 Mbits/sec
[  6]  0.0-10.0 sec   125 MBytes   105 Mbits/sec
[  7]  0.0-10.0 sec   125 MBytes   105 Mbits/sec
[ 10]  0.0-10.0 sec   125 MBytes   105 Mbits/sec
[  8]  0.0-10.0 sec   125 MBytes   105 Mbits/sec
[  9]  0.0-10.0 sec   125 MBytes   105 Mbits/sec
[ 12]  0.0-10.0 sec  62.8 MBytes  52.5 Mbits/sec
[ 11]  0.0-10.0 sec  62.8 MBytes  52.5 Mbits/sec
[SUM]  0.0-10.0 sec  1.10 GBytes   941 Mbits/sec
odroid@odroid:~$ iperf -s
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size: 85.3 KByte (default)
------------------------------------------------------------
[  4] local 192.168.100.30 port 5001 connected with 192.168.100.2 port 50554
[  5] local 192.168.100.30 port 5001 connected with 192.168.100.2 port 50562
[  6] local 192.168.100.30 port 5001 connected with 192.168.100.2 port 50556
[  7] local 192.168.100.30 port 5001 connected with 192.168.100.2 port 50560
[  9] local 192.168.100.30 port 5001 connected with 192.168.100.2 port 50566
[  8] local 192.168.100.30 port 5001 connected with 192.168.100.2 port 50558
[ 10] local 192.168.100.30 port 5001 connected with 192.168.100.2 port 50564
[ ID] Interval       Transfer     Bandwidth
[  4]  0.0-10.1 sec   160 MBytes   134 Mbits/sec
[  6]  0.0-10.1 sec   160 MBytes   134 Mbits/sec
[  5]  0.0-10.1 sec   161 MBytes   134 Mbits/sec
[  7]  0.0-10.1 sec   160 MBytes   134 Mbits/sec
[  8]  0.0-10.1 sec   160 MBytes   134 Mbits/sec
[ 10]  0.0-10.1 sec   160 MBytes   133 Mbits/sec
[  9]  0.0-10.1 sec   160 MBytes   134 Mbits/sec
[SUM]  0.0-10.1 sec  1.10 GBytes   936 Mbits/sec
User avatar
Brian.K
 
Posts: 244
Joined: Tue Sep 15, 2015 7:30 pm
Location: South Korea
languages_spoken: English, Korean
ODROIDs: XU4, C1+, C2

Re: Stack-able micro computers

Unread postby Waffle » Tue Aug 22, 2017 2:39 pm

I provide MDXfind, a hash auditing tool. I've been building versions for ARM for a few years now, and was looking forward to trying it on the Odroid. I was quite impressed - over 2x faster than a PI3, and even faster than the Firefly RK3399, withe Cortex-A72 chip. The XU4 is running at more than 10M MD5's per second, using 8 cores.

MDXfind handles hundreds of hash formats, and is faster than programs like Hashcat for many tasks - and Hashcat doesn't run on the ARM :-)

I'm in the process of adding distributed support (via AES-encrypted TCP connections) to many machines, so the minimalist XU4S is a perfect choice for this. Great job, Hardkernel engineers!
Waffle
 
Posts: 1
Joined: Tue Aug 22, 2017 2:26 pm
languages_spoken: english
ODROIDs: 2xXU4

Re: Stack-able micro computers

Unread postby bronco » Tue Aug 22, 2017 2:41 pm

joshi wrote:1) From disk IO perspective - ~8 mbps write, at least 400 mbps read


Should work. With a good SSD and kernel 4.x (requirement to use UAS) we get +240MB/s write and +300 MB/s read (with sequential workloads using large blocksizes). With smaller block sizes it depends. Test done with two OS images using either 3.10 or 4.9 kernel and with both their defaults and after switching to performance governor: https://pastebin.com/i1SQGAji (test SSD was a Samsung EVO 840 120 GB -- the slowest EVO available and this iozone call:
Code: Select all
iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2


Thank you for iperf numbers @Brian.K -- there' still something wrong with my network here since performance in one direction is only half of what's possible.
bronco
 
Posts: 27
Joined: Tue Jul 11, 2017 2:58 pm
languages_spoken: english

Re: Stack-able micro computers

Unread postby felixhermann » Tue Aug 22, 2017 4:21 pm

hello forum

I would be interested in testing an MC-1 cluster. Last year I successfully built a raspi cluster and use it for cfd-calculations with up to 1000000 gridpoints. the cluster runs on ubuntu. please let me know if you are still looking for testing people

best regards
felix
felixhermann
 
Posts: 1
Joined: Tue Aug 22, 2017 4:13 pm
languages_spoken: english
ODROIDs: none yet

Re: Stack-able micro computers

Unread postby ard » Tue Aug 22, 2017 7:15 pm

joshi wrote:Have anybody tried this as a streaming device? I don't think it's advisable to use RTSP or RTMP on such setup, but HLS or DASH might be quite nice given you have a decent SSD.
The question is - will it handle streaming load:
1) From disk IO perspective - ~8 mbps write, at least 400 mbps read
2) From network connection perspective - 1 incoming UDP multicast, at least 50 outgoing HTTP

I guess only experiment might show.

On the other hand, in more balanced, home streaming device setup, it might work better, as the needs are different:
1) From disk IO perspective - ~70 mbps write, ~32 mbps read
2) From network connection perspective - either 1 incoming UDP multicast if demux is done locally or up to 20 of them if done externally and 4-5 outgoing HTTP ones.

I am not really sure what you are trying to ask.
I have an lxc on a xu4 in a cloudshell with harddisk, and it multicasts 4 ts-over-rtp streams. On the other hand I have a bunch of Xu4's that can real time software decode the 4 streams, and downscale them to 4 thumbs, or show 1 fullscreen (per xu4... It's awesome to see that many display the same thing with audio *in sync* across all the systems).
That same xu4 that does the multicasting also contains in another lxc and vlan my VoIP server. And collectd and munin. And 3 compile environments (ubuntu 14.04, ubuntu 16.04 and debian jessie).
I have yet to hear a frame drop on the VoIP server, but I guess I am not a guy that calls much :-).
And to be clear: it's the VoIP server for my mobile phone... The endpoint of my gsm is my voip server, and the endpoint of my mobile number is that same server 8-D.
ard
 
Posts: 55
Joined: Tue Jul 09, 2013 2:12 am
languages_spoken: english, dutch, german
ODROIDs: ODROID-U2

Re: Stack-able micro computers

Unread postby mad_ady » Tue Aug 22, 2017 9:46 pm

@ard: Maybe you could write a nice article about it :) I'm interested in the VoIP gateway. And the streaming (purpose, volume).
User avatar
mad_ady
 
Posts: 2400
Joined: Wed Jul 15, 2015 5:00 pm
Location: Bucharest, Romania
languages_spoken: english
ODROIDs: XU4, C1+, C2

Re: Stack-able micro computers

Unread postby joshi » Tue Aug 22, 2017 10:16 pm

ard wrote:I am not really sure what you are trying to ask.
I have an lxc on a xu4 in a cloudshell with harddisk, and it multicasts 4 ts-over-rtp streams.


Thanks, ard, that draws partial picture about capabilities of this machine for me. Can you please elaborate more on that streaming application?
1) Bitrates of those 4 streams?
2) Are they permanently stored on HDD or there are simultaneous write/read operations going on? I.e. do streams come from external device, being stored on HDD and streamed from arbitrary positions or this is more like VoD scenario, when streams are written once and streamed on demand?
joshi
 
Posts: 3
Joined: Tue Aug 22, 2017 6:06 am
languages_spoken: english, russian, georgian

Re: Stack-able micro computers

Unread postby joshi » Tue Aug 22, 2017 10:43 pm

bronco wrote:
joshi wrote:1) From disk IO perspective - ~8 mbps write, at least 400 mbps read


Should work. With a good SSD and kernel 4.x (requirement to use UAS) we get +240MB/s write and +300 MB/s read (with sequential workloads using large blocksizes). With smaller block sizes it depends. Test done with two OS images using either 3.10 or 4.9 kernel and with both their defaults and after switching to performance governor: https://pastebin.com/i1SQGAji (test SSD was a Samsung EVO 840 120 GB -- the slowest EVO available and this iozone call:
Code: Select all
iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2


Thank you for iperf numbers @Brian.K -- there' still something wrong with my network here since performance in one direction is only half of what's possible.


Thanks bronco,

Write operations will be sequential with adjustable block size, read operations are mostly chaotic and I guess I will have to spare half of the RAM for read cache as in most scenarios 80% of timeshifted content belongs to last 30 minutes and I can get a good cache hit ratio on that.

So I guess more realistic numbers would be 150 by 200 and it seems to suit home PVR appliance scenario more, with 1 transponder on input an 4-5 clients on output.
joshi
 
Posts: 3
Joined: Tue Aug 22, 2017 6:06 am
languages_spoken: english, russian, georgian

Re: Stack-able micro computers

Unread postby tkaiser » Wed Aug 23, 2017 4:26 pm

tkaiser
 
Posts: 95
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: Stack-able micro computers

Unread postby tkaiser » Fri Aug 25, 2017 6:05 pm

I wonder whether anyone with appropriate consumption measurement equipment already did some research on switching off unneeded Exynos IP blocks on HC1 (talking about HDMI and if possible GPU)?

I measured 3.7W idle with Hardkernel's 5V/4A PSU and a connected but sleeping HDD (see post #3 of the mini review above). While I don't trust that much in my Powermeter at least relative comparisons should be possible and with a setup providing the same NAS performance (ROCK64 + same HDD + JMS567 drive enclosure) my powermeter shows just 1.8W in idle (and this is prior to any optimizations like disabling IP blocks since with ROCK64 development currently focuses on getting things up and running -- disabling / power savings are postponed atm).

Edit: Silly me, my measurements with ROCK64 were totally wrong! The JMS567 enclosure was powered by another PSU (my usual setup for performance tests to prevent underpowering affecting benchmark numbers). As soon as I let ROCK64 power the enclosure + (still sleeping) HDD the powermeter shows 3.2W. After playing hot swap on the SATA connector (don't do this at home!) it's 2.6W (most probably JMS567's SATA PHY needs 600mW in active mode).

In other words: we're talking about just 0.5W difference between HC1 and a ROCK64 idling around with connected disk. This difference is negligible but it would be still interesting whether an own device tree file for HC1 shows further consumption savings or not.
tkaiser
 
Posts: 95
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: Stack-able micro computers

Unread postby MimCom » Fri Aug 25, 2017 11:24 pm

Have you considered supporting 802.3af Power Over Ethernet? PoE switches are quite inexpensive these days and would minimize cabling while also facilitating remote power-cycling of any node.
MimCom
 
Posts: 9
Joined: Sun Mar 12, 2017 3:24 am
languages_spoken: english

Re: Stack-able micro computers

Unread postby elatllat » Fri Aug 25, 2017 11:44 pm

tkaiser wrote:... HC1 ... W ... vs ... ROCK64 ... W ...

They are both 28 nm lithography so I would expect similar power consumption.
Ethernet I would guess would be the largest power consumer of the peripherals.
ROCK64 has USB3 like the XU4 and claims 2160p@60hz@10bit which is better than the C2 hardware wise but the ROCK64 has no software to fully use 4k yet.
If you want more power savings use wifi instead of Ethernet and a 10nm chip, and a board with a standby mode... aka install Linux (Debian no-root) on a Samsung Galaxy S8 or OnePlus 5 and disable the screen... that's costly so we can only ask Odroid to snuggle up to Samsung or Qualcomm and make an SBC with the Exynos 9 Octa 8895 or Snapdragon 835.
Last edited by elatllat on Sat Aug 26, 2017 2:16 am, edited 1 time in total.
elatllat
 
Posts: 561
Joined: Tue Sep 01, 2015 8:54 am
languages_spoken: english
ODROIDs: XU4

Re: Stack-able micro computers

Unread postby linuxest » Sat Aug 26, 2017 2:04 am

I received my HC1 this morning. It works nicely with the latest Ubuntu minimal image based on kernel 4.9.44.
I plugged a cheapy WD 240GB Green SSD and moved the rootfs from SD card to the SSD with a rsync command.
The SSD is amazingly fast.. access spees is very close to 300MB/s. :o Clean kernel build took less than 20 minutes with the SSD. I love this snappy little machine.
The idle power consumption is 3.5Watt with stock performance governor. It is 2.9Watt now once I swiched to ondemand.
I'll play with apache, nginx, php and mysql packages this weekend to replace my 10 years old ~150Watt server.
I couldn't predict we can have a very powerful palm-sized server today. It is really affordable.
linuxest
 
Posts: 31
Joined: Tue Nov 08, 2016 6:35 pm
languages_spoken: english
ODROIDs: 1 x HC1 and 2 x C2s with HiFi-Shield 2
And some RPi3 boards

Re: Stack-able micro computers

Unread postby tkaiser » Sat Aug 26, 2017 5:21 pm

elatllat wrote:
tkaiser wrote:... HC1 ... W ... vs ... ROCK64 ... W ...

They are both 28 nm lithography so I would expect similar power consumption.
Ethernet I would guess would be the largest power consumer of the peripherals.


Great. I was not talking about lithography, future products or other irrelevant stuff but about the try to optimize settings for an already existing device called HC1 where we can't make use of HDMI (and most users don't want to use the Mali GPU too since OpenCL use cases are not that common). Based on tests with other SoCs we know that disabling certain IP blocks yields some consumption savings. Sometimes this is done by the driver (eg. checking for a connected HDMI display and deactivating HDMI PHY after 10 minutes of inactivity), sometimes this has to be done manually.

On HC1 IMO it's worth a try to test with HDMI and Mali GPU but my naive tries below yielded no (significant) savings (maybe 50mW according to powermeter) but as already said I lack the equipment to measure more precise with XU4/HC1).

Code: Select all
root@odroidxu4:/boot/dtb# diff exynos5422-odroidxu4.dts exynos5422-odroidxu4-original.dts
9c9
<    model = "Hardkernel Odroid XU4 mod";
---
>    model = "Hardkernel Odroid XU4";
2257c2257
<          status = "disabled";
---
>          status = "okay";
3780c3780
<       status = "disabled";
---
>       status = "okay";


BTW: HC1 Ethernet consumption and relationship with clockspeeds (and relationship with IRQ affinity) is already well known since tested: https://forum.armbian.com/index.php?/to ... ment=37989 and you will be surprised that a ROCK64 under full load consumes only half as much as a HC1 for obvious reasons :)
tkaiser
 
Posts: 95
Joined: Mon Nov 09, 2015 12:30 am
languages_spoken: english
ODROIDs: C1+, C2, XU4, HC1

Re: Stack-able micro computers

Unread postby MrDreamBot » Sun Aug 27, 2017 8:55 am

Image
Just received my ODROID-MC1, I did not realise it is that compact until I opened the box. The above is the size comparison of the MC1 to the VuShall and the CloudShell 1.
And here is the makeshift setup of the MC1 with a Gigabit switch.
Image
I am happy to report that the MC1 runs docker 1.12.6 without problem. I shall do some more testing later. :lol:
User avatar
MrDreamBot
 
Posts: 24
Joined: Fri Mar 13, 2015 10:58 am
languages_spoken: english
ODROIDs: ODROID-C2, ODROID-C1/C1+, ODROID-XU3, ODROID-XU4, ODROID-MC1, ODROID-HC1,ODROID-U3

Re: Stack-able micro computers

Unread postby moon.linux » Sun Aug 27, 2017 11:47 am

I was thinking that MC1 would be a extension on HC1 with support of sata driver on each nodes, but this looks different. :shock:
moon.linux
 
Posts: 851
Joined: Thu Oct 02, 2014 11:42 pm
languages_spoken: english

Re: Stack-able micro computers

Unread postby MrDreamBot » Sun Aug 27, 2017 3:12 pm

The circuit boards on the MC1 do not have a SATA interface like the HC1 as can be seen from the photo. The HC1 can be used as a NFS server for the MC1. I don't even know if the MC1 can be mounted on top of the HC1. Can odroid, the site admin, please clarify?
User avatar
MrDreamBot
 
Posts: 24
Joined: Fri Mar 13, 2015 10:58 am
languages_spoken: english
ODROIDs: ODROID-C2, ODROID-C1/C1+, ODROID-XU3, ODROID-XU4, ODROID-MC1, ODROID-HC1,ODROID-U3

PreviousNext

Return to News

Who is online

Users browsing this forum: No registered users and 1 guest