Releasing Android application for ARM processors
log in

Advanced search

Message boards : News : Releasing Android application for ARM processors

Previous · 1 · 2 · 3 · Next
Author Message
VictordeHollander
Send message
Joined: 15 Feb 13
Posts: 5
Credit: 1,327,560
RAC: 0
Message 981 - Posted: 11 Mar 2013, 22:39:27 UTC

HTC Desire HD (which is quite an old phone, october 2010)
ARM chip: Qualcomm Snapdragon S2 MSM8255 @1.0 GHz
68,749 sec = ~19 hours

Not very fast, but it uses very little power, something like 2-3 Watt.
So 19hours * 0.003KW = 0.057KWh for a WU

Compared to my 2500k @4.0 GHz that takes 1.5 hours, but it can do 4WUs at the same time. I estimate it uses something like 150W.
1.5hours * 0.150KW / 4WUs = 0.056KWh for a WU

In terms of energy efficiency they are roughly comparable. I think new smart-phones with dual or quad cores (@ higher GHz) will probably be even more energy efficient. On the other hand, Ivy Bridge is also ~20% more energy efficient than Sandy Bridge.

alex22
Send message
Joined: 15 Mar 13
Posts: 2
Credit: 480
RAC: 0
Message 987 - Posted: 15 Mar 2013, 18:50:01 UTC

Thanks Kyong for supporting ARM crunching.

Does this ARM application uses the GPU?
It goes amazingly fast on my first WU: 50% of 315000GFLOPS in 5.5h for one core. This is 10x faster than MilkyWay@Home on my:

Samsung Galaxy Note 2
ARMv7 4cores@57C, 1.6GHz@1150mV (undervoltage), 4 GPU@undervoltage
Governor: Performance
Android 4.1.2
Kernel Perseus 3.0.36
NativeBoinc

I'm happy to help your Asteroid research projects.

matszpk
Send message
Joined: 21 Jul 12
Posts: 7
Credit: 3,911,900
RAC: 362
Message 988 - Posted: 15 Mar 2013, 20:38:26 UTC

No. This version has been compiled for ARM with VFP. Time may be comparable with MilkyWay WUs.

Profile Ray_GTI-R
Send message
Joined: 15 Jul 12
Posts: 11
Credit: 8,520
RAC: 0
Message 995 - Posted: 17 Mar 2013, 16:58:09 UTC

Hi.

Just attached some ARM devices. all Android devices running 100% CPU usage, standard clocks.

1 x ODROID U2/Android/NativeBOINC ...
.. 4 concurrent tasks, after 30 minutes:-
... 2.187%
... 4.496%
... 0.000% >>> ??? (1.823% after +/- 40 minutes then started motoring)
... 2.187%
All completed in +/- 9 hours 20 minutes.

1 x cheap generic pad/Android/NativeBOINC, 1 task, after 30 minutes 0.000% >>> ??? (3.852% after 5 hrs*)
Will probably take 6 days to complete.

And for good measure an RPi running BOINC @ 90% CPU (100% CPU always fails) under Raspbian and o/c to 900mhz etc, 1 task, after 30 minutes 0.000% then after 11 hours it shows 10% completed.
This also will probably take 6 days to complete.

HTH, Ray

* by comparison after 5 hours each concurrent U2 task was +50%. Which is nice.

alex22
Send message
Joined: 15 Mar 13
Posts: 2
Credit: 480
RAC: 0
Message 997 - Posted: 18 Mar 2013, 12:28:57 UTC

Finaly job was slower and credits were similar to Milkyway as you said.

Profile Ray_GTI-R
Send message
Joined: 15 Jul 12
Posts: 11
Credit: 8,520
RAC: 0
Message 999 - Posted: 20 Mar 2013, 23:42:46 UTC - in response to Message 995.

OK, both arm v6 devices completed slightly ahead of my rough calculations at about 4.6 days each task.
Compare with the ODROID U2 arm v7 that completes 4 tasks concurrently in about 9.3 hours.
Now I know the arm v6 isn't as powerful as the arm v7 (MHz, architecture etc).
But simple math can't explain why there is such a vast gulf in task duration between these ARM devices so I reckon the problem is RAM - in particular the lack of onboard cache built into the arm v6 devices. For example, my model RPi has only 256mb compared to the 2gb of the arm v7 shared by 4 cores.

To use an old Windows analogy it looks like the arm v6 devices are "thrashing". I think the tiny memory in the arm v6 devices simply can't cope with the demands placed by Asteroids and dump to the (slower) swapfile, have to re-read it, dump, re-read etc etc.

Other BOINC projects work much more happily on the arm v6 devices. For example Subsetsum completes in hours rather than days.

Does anyone know a simple way of checking if this "thrashing" is taking place? FWIW my RPi runs Raspbian.

TIA, Ray

matszpk
Send message
Joined: 21 Jul 12
Posts: 7
Credit: 3,911,900
RAC: 362
Message 1000 - Posted: 21 Mar 2013, 12:43:46 UTC

This app was not optimized in the assembly (this is only recompilation for ARM CPU). ARMv6 CPUs has less powerful FPU and cant execute code efficiently. To Ray_GTI-R: Maybe your the ARMv6 hardware is really slow ?? Others ARMv6 devices computes workunits in about 2-3 days.

Profile Ray_GTI-R
Send message
Joined: 15 Jul 12
Posts: 11
Credit: 8,520
RAC: 0
Message 1016 - Posted: 30 Mar 2013, 1:22:00 UTC - in response to Message 1000.

Hi matszpk
Yes, my ARMv6 devices are slow (Raspberry Pi @ 900 mhz & Disgo 7000). Really, really slow.

Jonathan Rodriguez
Send message
Joined: 13 Mar 13
Posts: 4
Credit: 15,360
RAC: 0
Message 1026 - Posted: 3 Apr 2013, 18:50:33 UTC - in response to Message 953.

sign me up, im very interseted

Jonathan Rodriguez
Send message
Joined: 13 Mar 13
Posts: 4
Credit: 15,360
RAC: 0
Message 1027 - Posted: 3 Apr 2013, 18:50:37 UTC - in response to Message 953.

sign me up, im very interseted

osoyoosking
Send message
Joined: 26 Jul 12
Posts: 2
Credit: 354,480
RAC: 0
Message 1028 - Posted: 5 Apr 2013, 0:17:56 UTC - in response to Message 953.

hi just wanted to post i set up my Samsung S3 with nativeboinc since i use the pc ver with alot of hostes why not cell phones i have around the house too :)) so far so good.. my phone has duel core so will be cool to see what it puts out.

Profile Pooh Bear 27
Avatar
Send message
Joined: 3 Apr 13
Posts: 29
Credit: 1,019,280
RAC: 0
Message 1029 - Posted: 5 Apr 2013, 12:55:15 UTC

I crunched my first two units on my phone, and my results were not very typical of other ARM devices. I do not see this same issue on other projects.

For example
Run time 115,292.95
CPU time 56,564.83

Run time is over double CPU time.

It also never gave up the slots during the whole crunching time. I had two other units from another project, but the switch between never happened. In fact after the first two finished, two more downloaded and locked up both slots again leaving the two others suspended. I decided to hand suspend Asteroids until the other projects finished. I will see how these next two process without the other project waiting.

Any help in why I would have such an outrageous run tin would be appreciated,

osoyoosking
Send message
Joined: 26 Jul 12
Posts: 2
Credit: 354,480
RAC: 0
Message 1045 - Posted: 7 Apr 2013, 13:31:37 UTC - in response to Message 1028.

well been running couple days now and its working fine for me i sysnc with BAM and i even got my other list that i have on my 7 computers and it only gets work that is able to us my armv7 @1500 it really cool program runs just like the pc ver.

Detailed stats
ARMv7 Processor rev 0 (v7l) @1512MHz
OverviewBest ten daysProject listChartsLast 40 days
Project name Current Credit % of total Credit
/day Credit
/week Credit
/month World Position last
day last
week last
month
BOINC combined 682.61 100.00 0 0 0 5,575,680 0 0 0
MilkyWay@home 213.76 31.31 214 214 214 291,083 0 0 0
yoyo@home 140.89 20.64 141 141 141 50,370 0 0 0
SubsetSum@Home 182.01 26.66 91 182 182 3,506 186 186 186
WUProp@Home 84.00 12.31 56 84 84 30,959 2,090 2,090 2,090
theSkyNet POGS 61.95 9.08 0 62 62 4,989 10 10 10

Retired Projects:
Project name Credit % of total World Position



theres what im getting S3@1500 run 24 hours a day last 3 days

undertech
Send message
Joined: 22 Mar 13
Posts: 3
Credit: 1,560
RAC: 0
Message 1054 - Posted: 10 Apr 2013, 19:12:33 UTC

started running this project on my Galaxy Tab, and the first WU errored out ...

http://asteroidsathome.net/boinc/workunit.php?wuid=1476643 -> this is the WU i am talking about

CPU is running between 800Mhz and 1.2 Ghz. Network connection isn't always present, and in the messages tab it says that the task exited with 0 status... is this a bug of the app, or is it my fault?

fractal
Send message
Joined: 12 Jul 12
Posts: 5
Credit: 19,021,195
RAC: 0
Message 1056 - Posted: 11 Apr 2013, 6:08:53 UTC - in response to Message 1054.

Are you running openboinc? I run that on my tablet and it errored out every work unit until I found I have to keep it connected to one and only one network 24x7 or it aborts the work. For me this meant telling the tablet to "forget" the second network in my house as it was switching between them for unknown reasons which caused the work in progress to abort.

Profile Edd Vanderpool
Avatar
Send message
Joined: 18 Mar 13
Posts: 1
Credit: 6,960
RAC: 0
Message 1078 - Posted: 12 Apr 2013, 15:52:39 UTC - in response to Message 1029.

same exp. with a couple of programs but I have only been around since the first week of April, so theres a lot yet to leqrn in my case. Thank You..

undertech
Send message
Joined: 22 Mar 13
Posts: 3
Credit: 1,560
RAC: 0
Message 1106 - Posted: 14 Apr 2013, 20:59:30 UTC - in response to Message 1056.

Are you running openboinc? I run that on my tablet and it errored out every work unit until I found I have to keep it connected to one and only one network 24x7 or it aborts the work. For me this meant telling the tablet to "forget" the second network in my house as it was switching between them for unknown reasons which caused the work in progress to abort.


No, i'm running NativeBOINC, didn't even knew that there was an alternative to it. Is openboinc better?

Well, i used the tablet outside home when i was computing, maybe it was because of that? If so, it seems like this is the only project that requires persistent Internet connection... This tablet rarely leaves home, but no other project errored out with no connection. Does setting NativeBOINC to use wifi only correct the problem? Will test it out later, maybe... Meanwhile, i'll be waiting for another WU from this project.

VictordeHollander
Send message
Joined: 15 Feb 13
Posts: 5
Credit: 1,327,560
RAC: 0
Message 1133 - Posted: 17 Apr 2013, 13:12:48 UTC - in response to Message 999.

@Ray_GTI-R
My ODROID-U2 is much slower than yours, how did you configure yours?
It is taking 18 hours to finish 4 concurrently!

Mine is running the Android 4.1.1 (1.5 beta from the hardkernel website) with nativeboinc and No-frills CPU Control to prevent it from lowering clocks.

undertech
Send message
Joined: 22 Mar 13
Posts: 3
Credit: 1,560
RAC: 0
Message 1144 - Posted: 22 Apr 2013, 19:15:28 UTC

Got another WU from this project, tablet was connected to wifi throughout it's crunching, and no problems whatsoever while crunching... Seems that this project really needs persistent web connection.

Meanwhile, I was trying to report this WU, but apparently the project is down... So I still don't know if this WU will be validated.

Profile Michael McMahon
Avatar
Send message
Joined: 19 Jun 12
Posts: 2
Credit: 199,680
RAC: 108
Message 1438 - Posted: 2 Aug 2013, 16:40:39 UTC

I don't know how I missed this notice, but I found the app when I was notified about POGS starting to go on Android. I am running the two alternating projects, and it seems to work good! Awesome job guys (and of course gals) :)
____________

Previous · 1 · 2 · 3 · Next
Post to thread

Message boards : News : Releasing Android application for ARM processors


Main page · Your account · Message boards


Copyright © 2019 Asteroids@home