amuck-landowner

ARM device with decent disk performance?

drmike

100% Tier-1 Gogent
Here is my speed test on a new Octa Core ARM box.  It has 32GB of disk built into the unit + has USB 3.0 + SATA...


root@allwinner:/home/linaro# hdparm -tT --direct /dev/mmcblk0

/dev/mmcblk0:
Timing O_DIRECT cached reads: 68 MB in 2.05 seconds = 33.13 MB/sec
Timing O_DIRECT disk reads: 98 MB in 3.03 seconds = 32.37 MB/sec



Code:
root@allwinner:/home/linaro# dd if=/dev/zero of=test bs=16k count=8k conv=fdatasync && rm -rf test
8192+0 records in
8192+0 records out
134217728 bytes (134 MB) copied, 4.16864 s, 32.2 MB/s




Not bad.  From the looks of it, the config as-is isn't too optimal either... probably can get more out of this.


Link for this device: http://www.rikomagic.com/en/product/showpro_id_70_pid_22.html
 

willie

Active Member
Wow, that looks like a nice box if a little bit power hungry.   Could you run:

  • openssl speed rsa1024 (scaleway: 196.4 sign, 3465.5 verify)
  • time echo "scale = 1000; 4*a(1)" | bc -l  (scaleway: 1.78 sec)
  • time echo "scale = 2000; 4*a(1)" | bc -l (scaleway: 10.28 sec)

Might ask for a few other tests involving downloads, if you're up for that.  You might have to apt-get install bc for the bc tests.


Can I ask where you bought the box, and what it cost?


Thanks!
 
Last edited by a moderator:

wlanboy

Content Contributer
  • openssl speed rsa1024 (scaleway: 196.4 sign, 3465.5 verify)
  • time echo "scale = 1000; 4*a(1)" | bc -l  (scaleway: 1.78 sec)
  • time echo "scale = 2000; 4*a(1)" | bc -l (scaleway: 10.28 sec)

A20-OLinuXino-LIME:

  • openssl speed rsa1024
Code:
openssl speed rsa1024
Doing 1024 bit private rsa's for 10s: 1617 1024 bit private RSA's in 9.98s
Doing 1024 bit public rsa's for 10s: 29040 1024 bit public RSA's in 9.98s
OpenSSL 1.0.1e 11 Feb 2013
built on: Thu Dec  3 18:30:13 UTC 2015
options:bn(64,32) rc4(ptr,char) des(idx,cisc,16,long) aes(partial) blowfish(ptr)
compiler: gcc -fPIC -DOPENSSL_PIC -DZLIB -DOPENSSL_THREADS -D_REENTRANT -DDSO_DLFCN -DHAVE_DLFCN_H -DL_ENDIAN -DTERMIO -g -O2 -fstack-protector --param=ssp-buffer-size=4 -Wformat -Werror=format-security -D_FORTIFY_SOURCE=2 -Wl,-z,relro -Wa,--noexecstack -Wall -DOPENSSL_BN_ASM_MONT -DOPENSSL_BN_ASM_GF2m -DSHA1_ASM -DSHA256_ASM -DSHA512_ASM -DAES_ASM -DGHASH_ASM
                  sign    verify    sign/s verify/s
rsa 1024 bits 0.006172s 0.000344s    162.0   2909.8
  • time echo "scale = 1000; 4*a(1)" | bc -l
Code:
time echo "scale = 1000; 4*a(1)" | bc -l
3.141592653589793238462643383279502884197169399375105820974944592307\
81640628620899862803482534211706798214808651328230664709384460955058\
22317253594081284811174502841027019385211055596446229489549303819644\
28810975665933446128475648233786783165271201909145648566923460348610\
45432664821339360726024914127372458700660631558817488152092096282925\
40917153643678925903600113305305488204665213841469519415116094330572\
70365759591953092186117381932611793105118548074462379962749567351885\
75272489122793818301194912983367336244065664308602139494639522473719\
07021798609437027705392171762931767523846748184676694051320005681271\
45263560827785771342757789609173637178721468440901224953430146549585\
37105079227968925892354201995611212902196086403441815981362977477130\
99605187072113499999983729780499510597317328160963185950244594553469\
08302642522308253344685035261931188171010003137838752886587533208381\
42061717766914730359825349042875546873115956286388235378759375195778\
18577805321712268066130019278766111959092164201988

real    0m2.455s
user    0m2.420s
sys     0m0.010s
  • time echo "scale = 2000; 4*a(1)" | bc -l
Code:
time echo "scale = 2000; 4*a(1)" | bc -l
3.141592653589793238462643383279502884197169399375105820974944592307\
81640628620899862803482534211706798214808651328230664709384460955058\
22317253594081284811174502841027019385211055596446229489549303819644\
28810975665933446128475648233786783165271201909145648566923460348610\
45432664821339360726024914127372458700660631558817488152092096282925\
40917153643678925903600113305305488204665213841469519415116094330572\
70365759591953092186117381932611793105118548074462379962749567351885\
75272489122793818301194912983367336244065664308602139494639522473719\
07021798609437027705392171762931767523846748184676694051320005681271\
45263560827785771342757789609173637178721468440901224953430146549585\
37105079227968925892354201995611212902196086403441815981362977477130\
99605187072113499999983729780499510597317328160963185950244594553469\
08302642522308253344685035261931188171010003137838752886587533208381\
42061717766914730359825349042875546873115956286388235378759375195778\
18577805321712268066130019278766111959092164201989380952572010654858\
63278865936153381827968230301952035301852968995773622599413891249721\
77528347913151557485724245415069595082953311686172785588907509838175\
46374649393192550604009277016711390098488240128583616035637076601047\
10181942955596198946767837449448255379774726847104047534646208046684\
25906949129331367702898915210475216205696602405803815019351125338243\
00355876402474964732639141992726042699227967823547816360093417216412\
19924586315030286182974555706749838505494588586926995690927210797509\
30295532116534498720275596023648066549911988183479775356636980742654\
25278625518184175746728909777727938000816470600161452491921732172147\
72350141441973568548161361157352552133475741849468438523323907394143\
33454776241686251898356948556209921922218427255025425688767179049460\
16534668049886272327917860857843838279679766814541009538837863609506\
80064225125205117392984896084128488626945604241965285022210661186306\
74427862203919494504712371378696095636437191728746776465757396241389\
086583264599581339047802759008

real    0m14.080s
user    0m14.040s
sys     0m0.010s

Can't believe that Scaleway is that slow.
 

drmike

100% Tier-1 Gogent
Wow, that looks like a nice box if a little bit power hungry.   Could you run:

  • openssl speed rsa1024 (scaleway: 196.4 sign, 3465.5 verify)
  • time echo "scale = 1000; 4*a(1)" | bc -l  (scaleway: 1.78 sec)
  • time echo "scale = 2000; 4*a(1)" | bc -l (scaleway: 10.28 sec)

Might ask for a few other tests involving downloads, if you're up for that.  You might have to apt-get install bc for the bc tests.


Can I ask where you bought the box, and what it cost?


Thanks!

This box like many of these specs out nice on paper but fails to deliver in Linux as shipped by default.  Sure the upfront base disk speeds looked good. That's where it ends for real world though.


Each board seems to have it's own little special oddness.  Ditto for the CPUs on ones like this.


Although they call this an Octa Core it is and isn't.   It has 4 cores of one type at a higher MHz level and then 4 cores of some other type at a lower MHz level.  Haven't looked at or found CPU sheet to demystify it yet.   Linux shows 8 cores, but I fail to see them utilized yet.  I've seen 2 cores utilized with things I run to see if my stack will work find on the device. The cores work when I tested openssl not linear increase near proportional to the cores, but like I said I don't think they all clock in at the same.


While my stack runs on this device, the performance sucks and is wrong direction with default base install.  That said running OpenJDK + MySQL + scripts.  Page generation times on this device are basically 300ms.  Raspberry Pi quad core does 100ms on average faster.  Being outpaced by a Pi?  Meh.


I am blaming Linaro as being poured in, but not optimized at all for this.  That's what it shipped with, so I need to find support / hobbyists out there and figure out what OS works better / or optimizations.


Let's run your tests on my Odroid U2 to compare with Scaleway.  The U2 is old, the original Odroid model I do believe, still capable :)

  • openssl speed rsa1024
    odroid u2: 326.7 sign, 5844.5 verify
    (scaleway: 196.4 sign, 3465.5 verify)
  • time echo "scale = 1000; 4*a(1)" | bc -l 
    odroid u2: 1.152 sec
    (scaleway: 1.78 sec)
  • time echo "scale = 2000; 4*a(1)" | bc -l
    odroid u2: 6.479 sec
    (scaleway: 10.28 sec)

So on those, the Odroid U2 trounces the Scaleway.

  • openssl speed rsa1024
    (Rikomagic MK80LE: 442.5 sign, 7789.3 verfiy)
    (odroid u2: 326.7 sign, 5844.5 verify)
    (scaleway: 196.4 sign, 3465.5 verify)
     
  • time echo "scale = 1000; 4*a(1)" | bc -l 
    (Rikomagic MK80LE: .998 sec)
    (odroid u2: 1.152 sec)
    (scaleway: 1.78 sec)
     
  • time echo "scale = 2000; 4*a(1)" | bc -l
    (Rikomagic MK80LE: 5.489 sec)
    (odroid u2: 6.479 sec)
    (scaleway: 10.28 sec)

So the Rikomagic MK80LE beats out the Odroid on these and Scaleway runs last.


Now let's run this with the Raspberry Pi quad core with default board settings:

  • openssl speed rsa1024
    (Rapberry Pi Quad Core: 147 sign 2645.5 verify)
    (Rikomagic MK80LE: 442.5 sign, 7789.3 verfiy)
    (odroid u2: 326.7 sign, 5844.5 verify)
    (scaleway: 196.4 sign, 3465.5 verify)
     
  • time echo "scale = 1000; 4*a(1)" | bc -l 
    (Rapberry Pi Quad Core: 2.201 sec)
    (Rikomagic MK80LE .998 sec)
    (odroid u2: 1.152 sec)
    (scaleway: 1.78 sec)
     
  • time echo "scale = 2000; 4*a(1)" | bc -l
    (Rapberry Pi Quad Core: 12.518 sec)
    (Rikomagic MK80LE 5.489 sec)
    (odroid u2: 6.479 sec)
    (scaleway: 10.28 sec)

Based on these, the Raspberry Pi Quad Core is the slowest.


Since openssl supports multi core, lets run 4 cores on the Pi vs. the Rikomagic vs. Odroid U2

  • openssl speed rsa1024 -multi 4
    (Rapberry Pi Quad Core: 580.6 sign 10485.3 verify)
    (Rikomagic MK80LE: 1707.4 sign 29144.2 verify )
    (odroid u2: 1295.1 sign 23359.8 verify)

    Now since the Riko has 8 cores, we run multi 8 test
  • openssl speed rsa1024 -multi 8
    (Rikomagic MK80LE: 2467.6 sign 42483.2 verify )
 
Last edited by a moderator:

drmike

100% Tier-1 Gogent
I'd be wary of the Rikomagic devices.  Not the best track record of shipping software stack that utilizes the hardware.


Tronsmart has a similar device if you are interested in USB 3, SATA, larger RAM, etc.


Tronsmart Draco AW80 is the model you want to check out.


Ubuntu in one review for that model wasn't exactly entirely ready, but it appears to be ahead of the Rikomagic at this point compared to box I have.


Tronsmart is hit or miss.  Upside is their devices are easier to find the States and better prices.  I have a dual core settop box from them that I need to get back to using.  Decent device for Android.
 

drmike

100% Tier-1 Gogent
Figure I'd try the Ubuntu  installed on this Rikomagic.


YouTube HD just won't work.. horrendously overloaded.  Turn it down to 480 and it works, but cores all going freaking busy crazy.  So the video card and acceleration clearly aren't working as shipped.  CPU bound video.  MEH!


Searching for new software / fresh OS install howto.. ideally this is just old shipped stack.
 
Last edited by a moderator:

TheLinuxBug

New Member
Although they call this an Octa Core it is and isn't.   It has 4 cores of one type at a higher MHz level and then 4 cores of some other type at a lower MHz level.  Haven't looked at or found CPU sheet to demystify it yet.   Linux shows 8 cores, but I fail to see them utilized yet.  I've seen 2 cores utilized with things I run to see if my stack will work find on the device. The cores work when I tested openssl not linear increase near proportional to the cores, but like I said I don't think they all clock in at the same.

It it called big.LITTLE Technology

big.LITTLE Processing – How does it work?



The high performance and high efficiency CPU clusters are connected through a cache coherent interconnect fabric such as the ARM CoreLink™ CCI-400.This hardware coherency enables the same view of the memory to both the big and LITTLE CPU clusters. The processors look like one multicore CPU to the operating system (OS). User space software on a big.LITTLE SoC is identical to the software that would run on a standard Symmetrical Multi-Processing (SMP) CPU.


How does the work get scheduled to the right processor?



Global Task Scheduling (GTS) gives the OS awareness of the big and LITTLE processors, and the ability to schedule individual threads of execution on the appropriate CPU core based on dynamic run-time behavior. ARM has developed a kernel space patch set based on GTS called big.LITTLE MP that keeps track of load history as each thread runs, and uses the history to anticipate the performance needs of the thread next time it runs. This software is in operation on the use case graph above (Fig. 1) and in the measured results below (Fig. 2), where big.LITTLE technology is delivering energy savings as high as 75% for the same or higher delivered performance.

Cheers!
 
Last edited by a moderator:

drmike

100% Tier-1 Gogent
I am developing a crush on Orange Pi.  They have a 2GB of RAM version that might cover all my RAM needs .. and has SATA.


Check this out:


http://www.orangepi.org/downloadresources/


This folks is what support site should look like for these ARM boards.  Everything tidy in one spot.  Downloads to the current distro files.
 

wlanboy

Content Contributer
Figure I'd try the Ubuntu  installed on this Rikomagic.


YouTube HD just won't work.. horrendously overloaded. 


Searching for new software / fresh OS install howto.. ideally this is just old shipped stack.

Binary non-open super-closed packed with old kernel video drivers... welcome to the Allwinner SoC world. Hopefully some kernel hackers do buy your stuff...


My OlinuXino does have video drivers as binary package. FullHD working like a charm. 
 

TheLinuxBug

New Member
I am developing a crush on Orange Pi.  They have a 2GB of RAM version that might cover all my RAM needs .. and has SATA.


Check this out:


http://www.orangepi.org/downloadresources/


This folks is what support site should look like for these ARM boards.  Everything tidy in one spot.  Downloads to the current distro files.

I want to be excited for you, but the people doing the real work to get these boards mainstream is http://linux-sunxi.org They are the ones who are working to get it so you can run something other than the old 3.x linux versions with an old outdated kernel.  If you want to know where to find the best information about your Allwinner SoC linux-sunxi is the answer, both their webpage and IRC channel can be quite useful.  A lot of people give their time freely to help create and reverse engineer drivers which Allwinner will not.


Not saying they don't provide some good material for their board, but you will find most of the advancements being had for the H3 are being done by the linux-sunxi team.


my 2 cents.


Cheers!
 
Last edited by a moderator:

drmike

100% Tier-1 Gogent
Not saying they don't provide some good material for their board, but you will find most of the advancements being had for the H3 are being done by the linux-sunxi team.


my 2 cents.


Cheers!

Allwinner is clearly a PITA.  Quite a shame because they have capable gear.  Strange approach they are taking to not support their hardware for Linux and closed / lacking drivers.


The Sunxi site is a great reference.  Much appreciated on that.
 

drmike

100% Tier-1 Gogent
Can I ask where you bought the box, and what it cost?

I actually found the Rikomagic as an open box / used thing.  Paid very little for it relative to the upwards of $145-200 they seem to be asking for them currently.


I'll only buy ARM at $0 to $80.  Nothing shipping at this time to justify more unless they include a display or something original.
 

drmike

100% Tier-1 Gogent

Definitely an interesting device.   I've stayed away from that brand due to high price point relatively speaking and the OS option front side is OpenElec and Android.  Unsure if these have more community support out there.


I know the Gigabit on these is half speed / not up to spec due to shared or limited bus.  Tired of that limitation all over town...


Armbian has support for this board:


http://www.armbian.com/cubox-i/
 
Top
amuck-landowner