Re: [beagleboard] Java Benchmark on the BeagleBone Black

[Maxim Podbereznyy]

Any surprises? I fully agree about apples and oranges but imagine how much you can do by powering the $45 board from a few AA batteries and now imagine the super-duper giant i7 that requires a Nuclear Power station from one side and a 3000 RPM fan on another to chill it out :)

17.05.2013 23:18 пользователь "Tom Maynard" <tom....@gmail.com> написал:

I was interested in comparing the raw Java capability on my three platforms (Win/Lin on amd64, and Angstrom on BBB) so I rather randomly selected the SciMark 2.0 benchmark, not because it runs the gamut of Java capabilities, but primarily because it comes from the NIST -- an origin in which I place a reasonable amount of trust. And it flexes muscles with which I am at least familiar. I set my watch per the recommendation of the NIST, after all.... Their download page.

My query was simply: "How good is the BeagleBone Black, really?" I have run the Linux (my current host system) and BBB trials so far, with the Win7 test to be subsequent to my next reboot there.

After I ran the tests I immediately realized that I am incapable of analyzing the results, but as a pure comparison they may have some tiny value -- but no attempt was made to "level the playing field" in any way. Nothing is the same across systems. These are apples and oranges ... from different planets.

TL;DR The executive summary: Host system = 1,200, BBB = 12.

Intel Core i7 6GB Lubuntu 13.04 (the LXDE fork of Ubuntu):

╭─tom@venus  ~/Downloads  
╰─$ uname -a && cat /etc/*release
Linux venus 3.8.0-21-generic #32-Ubuntu SMP Tue May 14 22:16:46 UTC 2013 x86_64 x86_64 x86_64 GNU/Linux
DISTRIB_ID=Ubuntu
DISTRIB_RELEASE=13.04
DISTRIB_CODENAME=raring
DISTRIB_DESCRIPTION="Ubuntu 13.04"
NAME="Ubuntu"
VERSION="13.04, Raring Ringtail"
ID=ubuntu
ID_LIKE=debian
PRETTY_NAME="Ubuntu 13.04"
VERSION_ID="13.04"
HOME_URL="http://www.ubuntu.com/"
SUPPORT_URL="http://help.ubuntu.com/"
BUG_REPORT_URL="http://bugs.launchpad.net/ubuntu/"

╭─tom@venus  ~/Downloads  
╰─$ ls *.jar
scimark2lib.jar

╭─tom@venus  ~/Downloads  
╰─$ java -cp scimark2lib.jar jnt.scimark2.commandline

SciMark 2.0a

Composite Score: 1265.6632256172502
FFT (1024): 845.4874797092333
SOR (100x100):   1214.9779632490277
Monte Carlo : 658.5352790565404
Sparse matmult (N=1000, nz=5000): 1199.1948168457877
LU (100x100): 2410.1205892256617

java.vendor: Oracle Corporation
java.version: 1.7.0_21
os.arch: amd64
os.name: Linux
os.version: 3.8.0-21-generic

BeagleBone Black Ångström

╭─tom@venus  ~/Downloads  
╰─$ sftp root@beagle
root@beagle's password: 
Connected to beagle.
sftp> put scimark2lib.jar 
Uploading scimark2lib.jar to /home/root/scimark2lib.jar
scimark2lib.jar                               100%   33KB  33.4KB/s   00:00    
sftp> quit

╭─tom@venus  ~/Downloads  
╰─$ ssh root@beagle
root@beagle's password: 

╭─root@beaglebone ~
╰─# uname -a && cat /etc/*release
 Linux beaglebone 3.8.6 #1 SMP Sat Apr 13 09:10:52 CEST 2013 armv7l GNU/Linux
 NAME=Angstrom
 ID=angstrom
 PRETTY_NAME=The Ångström Distribution

╭─root@beaglebone ~
╰─# java -cp scimark2lib.jar jnt.scimark2.commandline

SciMark 2.0a

Composite Score: 12.321033912806325
FFT (1024): 8.048954713568516
SOR (100x100):   20.032383748321344
Monte Carlo : 5.7852467063020265
Sparse matmult (N=1000, nz=5000): 13.731142960110086
LU (100x100): 14.00744143572965

java.vendor: Sun Microsystems Inc.
java.version: 1.6.0_24
os.arch: arm
os.name: Linux
os.version: 3.8.6

╭─root@beaglebone ~
╰─#

--
For more options, visit http://beagleboard.org/discuss
---
You received this message because you are subscribed to the Google Groups "BeagleBoard" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beagleboard...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

[Tom Maynard]

On Friday, May 17, 2013 3:15:22 PM UTC-5, lisarden wrote:

Any surprises?

None: Everything just worked, right out of the box.  I spent more time getting Java installed than anything else.

 

imagine how much you can do by powering the $45 board from a few AA batteries

Oh, I have no complaints about the BBB -- it's a marvel that continues to amaze and delight me.  It's the most fun I've ever had for USD 50, and the fun is just beginning.

I did need something to compare the results with, and my laptop was just about my only choice.  Still, only two orders of magnitude difference is also amazing.  Hopefully one of my RasPi friends will run the same benchmark for a more meaningful comparison.

Gotta love the BBB!

[Maxim Podbereznyy]

Tom, RasPi should be slower but what it can do is 1080p decoding. Bbb just is not capable

18.05.2013 9:01 пользователь "Tom Maynard" <tom....@gmail.com> написал:

--

[Jason Stapels]

Just out of curiosity, what version of Java did you use and was it compiled using the VFP extensions?

[Tom Maynard]

On Saturday, May 18, 2013 10:11:03 PM UTC-5, Tom Maynard wrote:

On Saturday, May 18, 2013 8:26:48 PM UTC-5, Jason Stapels wrote:

Just out of curiosity, what version of Java did you use and was it compiled using the VFP extensions?

Assuming you mean "on the Beagle" the answer is:

java version "1.6.0_24"

OpenJDK Runtime Environment (IcedTea6 1.12pre) (6b24-1.11.1+1.11.2-devel+hg1+54ceda20a02c)

OpenJDK Zero VM (build 20.0-b12, mixed mode)

Since I don't know what "VFP extensions" are, I hope that answered your question. 

[Laurent H.]

The OpenJDK  ZeroVM is very slow because this is the "standard" OpenJDK without any assembly code ...

The Oracle VM contains some ARM assembly code that is not Open sourced. 

I ran some test on Raspberry Pi and Beaglebone (1st gen) : http://parleys.com/play/5156c4d6e4b0c779d7881405/chapter24/about / http://parleys.com/play/5156c4d6e4b0c779d7881405/chapter25/about

You can expect that the Oracle VM is about 4 to 5 times faster than OpenJDK ZeroVM.

[Tom Maynard]

On Monday, May 20, 2013 3:43:04 PM UTC-5, Laurent H. wrote:

The OpenJDK  ZeroVM is very slow because this is the "standard" OpenJDK without any assembly code ...

The Oracle VM contains some ARM assembly code that is not Open sourced. 

I'm certain that is the case, and I do have another Angstrom image with the Oracle JDK installed -- but I never found a real "opkg" package for the Oracle JDK, and thus while the code is there, and it works, the system doesn't "know" about it, and I have to spell out the pathnames (or adjust my PATH) to use it.

The OpenJDK is at least easier to obtain and install (especially for the average user).  The Angstrom "update-alternatives" only has an "install" option, no more!  I need the "set" option to make the Oracle JDK the default.  I never could figure out how to do that.

Any tips on doing that -- or on successfully installing an alternate distro -- would be very much appreciated.

[Nathaniel Lewis]

And you probably will never find an opkg for Oracle JDK.  They are very annoying when it comes to redistribution.  But its not so hard to use the Oracle JDK.  I just uninstalled the OpenJDK and untarred the arm variant of the oracle jdk into /usr/java on my BBB.  Then just added the bin directories to my PATH.  Simple as that really.

Nathaniel Lewis

CTO at E1FTW Games

2nd year Undergraduate, Computer Science and Engineering

Project Manager for the Robotics Society at UC Merced

Undergraduate Lab Technician Intern at MESA Labs

(925) 961 7578

linux.r...@gmail.com

[Tom Maynard]

On 05/21/2013 12:08 PM, Nathaniel Lewis wrote:

I just uninstalled the OpenJDK and untarred the arm variant of the oracle jdk into /usr/java on my BBB.  Then just added the bin directories to my PATH.  Simple as that really.

And so it was.  Since I have the luxury of space (running on a 16GB Class 10 HD card), I kept both JDKs for comparison purposes.  I have a simple shell script that switches between the two.

Thanks for the boost I needed to cross that hurdle (finally).

[Tim Miller]

I am curious, what OS are you running? And would that have an impact on your results?

-Tim

[Tom Maynard]

On Tuesday, June 4, 2013 5:17:42 AM UTC-5, Tim Miller wrote:

I am curious, what OS are you running? And would that have an impact on your results?

 

Well, on the BBB I'm running either the latest Angstrom (2013-05-27 or newer) or Robert C Nelson's Debian "Wheezy" build.  Those two are my primary boot images.

 

On the "big iron" I run Windows 8 Pro, or Ubuntu 13.04 -- also depending on which way I boot. 

 

I tend to experiment on the Debian boot -- mostly because I'm very familiar with Ubuntu/Debian-based systems -- and, if it works the way I want, I try to migrate the same thing to the Angstrom image.  Sometimes it works, sometimes it doesn't.   I also do almost everything at the command line, via SSH or "Shell in a Box," and only rarely hook up my Lapdock for GUI fun.

 

Curiosity cured?

[Tom Maynard]

On Tuesday, June 4, 2013 5:17:42 AM UTC-5, Tim Miller wrote:

I am curious, what OS are you running? And would that have an impact on your results?

 

Oops.  I forgot your Question #2: I really have no idea.  The underlying hardware is identical so you'd only be considering the innate latency of the OS ... and I can't begin to speculate what that might be.

 

I haven't done any comparisons, since that's not a consideration for me when selecting an SD card to boot from.  Try it, find out, and report back! 

[fe7...@gmail.com]

Hey,
I am interested in the open JDK java performance of the RaspberryPi and the BBB.
Does anyone have a source for binary packages of open JDK?
I used the yum package manager on the raspberry:
java version "1.7.0_b147-icedtea"
OpenJDK Runtime Environment (fedora-2.1.fc17.6-arm)
OpenJDK Zero VM (build 22.0-b10, interpreted mode)

And opkg on the BBB:

java version "1.6.0_24"
OpenJDK Runtime Environment (IcedTea6 1.12pre)

OpenJDK Zero VM (build 20.0-b12, mixed mode)

To be able to compare the systems it would be best to have a binary package that I can put on both systems so I have the exact same version.

I would appreciate your help
Felix

[Laurent H.]

You can use the Oracle JVM binaries : http://www.oracle.com/technetwork/java/javase/downloads/index.html

They both run on the Rpi and the BB.

I have made some benchmark tests showing that the JVM is faster on BB (first-gen) by a factor between 2 and 3.

Laurent