next iteration of Android OS will move away from Oracle's Java to OpenJDK

[Dabombinable]

Good thing Android is open source, right? XobotOS, aka Android rewritten in C#

Can it run normal APK, if so I've got an excellent smartphone that's already rooted....

[suicidalfranco]

Can it run normal APK, if so I've got an excellent smartphone that's already rooted....

don't know, never followed the development since it was announced

[Philosobyte]

Not extremely. It's between 1/2 and 1/3 the speed of good equivalent C++ code. Python is slow.

What's your source for this conclusion?

[patrickjp93]

What's your source for this conclusion?

In my HPC class we spend the first two weeks learning to do runtime analysis in Linux using /usr/bin/time -v

And then we analyze and report on about 8 different workloads with varying input sizes.

Pretty consistently good Java isn't any worse than 1/3 speed of good equivalent C++ under the GCC compiler. Under ICC it's between 1/3 and 1/4 as of spring last year.

[Philosobyte]

In my HPC class we spend the first two weeks learning to do runtime analysis in Linux using /usr/bin/time -v

And then we analyze and report on about 8 different workloads with varying input sizes.

Pretty consistently good Java isn't any worse than 1/3 speed of good equivalent C++ under the GCC compiler. Under ICC it's between 1/3 and 1/4 as of spring last year.

That's disappointing. I've invested a lot into Java for its cross-platform functionality, but I didn't think it was that much slower than C++

[patrickjp93]

That's disappointing. I've invested a lot into Java for its cross-platform functionality, but I didn't think it was that much slower than C++

C++ is as cross-platform as it gets unless you're using OS-specific interactions or libraries. The JVM is built from C++ for a reason. You'll never interpret and optimize code so quickly you'll do better than just writing it well and putting it on the hardware.

[Blade of Grass]

In my HPC class we spend the first two weeks learning to do runtime analysis in Linux using /usr/bin/time -v

And then we analyze and report on about 8 different workloads with varying input sizes.

Pretty consistently good Java isn't any worse than 1/3 speed of good equivalent C++ under the GCC compiler. Under ICC it's between 1/3 and 1/4 as of spring last year.

Are you saying that Java takes no more then 33% longer to run, or are you saying it can do no more then 33% of the operations over a period of time?

[patrickjp93]

Are you saying that Java takes no more then 33% longer to run, or are you saying it can do no more then 33% of the operations over a period of time?

 

In cases of high core saturation, Java can sometimes only do as good as 1/3 the work as C++ in the same period of time due to the overhead of the JVM.

[LukaP]

In cases of high core saturation, Java can sometimes only do as good as 1/3 the work as C++ in the same period of time due to the overhead of the JVM.

Do you have any sources on this? Because in my experience, algorithmically identical Java code will run from 1/4 to 1/8 the speed of C++... Granted i didnt do HPC with super optimised compilers, but more like low power apps and utilities but it ran at those speeds. Especially memory intensive tasks were slow as balls

[patrickjp93]

Do you have any sources on this? Because in my experience, algorithmically identical Java code will run from 1/4 to 1/8 the speed of C++... Granted i didnt do HPC with super optimised compilers, but more like low power apps and utilities but it ran at those speeds. Especially memory intensive tasks were slow as balls

 

I'll see if I can dig up my old HPC assignments. We did Diff. EQ. solving, parallel sorting, parallel searching, matrix multiplication, and a few obscure things, but depending on the task, it was about 1/3 speed at worst. Mind you that was with Core 2 Duo machines last year. The new kids get I7 4790s in all the lab machines .

[LukaP]

I'll see if I can dig up my old HPC assignments. We did Diff. EQ. solving, parallel sorting, parallel searching, matrix multiplication, and a few obscure things, but depending on the task, it was about 1/3 speed at worst. Mind you that was with Core 2 Duo machines last year. The new kids get I7 4790s in all the lab machines .

ill be glad to check it out ^^ 

oh fancy pantsies. we get i5 45XXs in classrooms. Though in our lab we have E7v3s

[patrickjp93]

ill be glad to check it out ^^ 

oh fancy pantsies. we get i5 45XXs in classrooms. Though in our lab we have E7v3s

 

When the Redhawk cluster upgrades to Skylake E7s at the start of next year, then we can brag

 

No need for E3 Xeons in a student lab.

[Sauron]

I can only be happy for wider adoption of open standards.

[LukaP]

When the Redhawk cluster upgrades to Skylake E7s at the start of next year, then we can brag

 

No need for E3 Xeons in a student lab.

well i hate the fact we have E7s as they are super expensive and not the best when it comes to memory bandwith. ive been lobying for Power8 all the time, but people there are used to x86 so oh well

 

Meh students should be required to have their own laptops. its not like they need the power of an i5... they could all live with an ultrabook for the stuff theyre working on...

[patrickjp93]

well i hate the fact we have E7s as they are super expensive and not the best when it comes to memory bandwith. ive been lobying for Power8 all the time, but people there are used to x86 so oh well

 

Meh students should be required to have their own laptops. its not like they need the power of an i5... they could all live with an ultrabook for the stuff theyre working on...

 

Skylake E7 brings on 6 x DDR4 2400, so bandwidth won't be an issue (it's socket-compatible with non-Omnipath KNL).

 

The non-engineering computer labs just use mobile-SKU NUCs. The CSE building also lends out its labs to the chemical, mechanical, electrical, and computer engineering students as a way to consolidate space, so they have all the power for CAD for the seniors and grad students, as well as the flexibility for use in HPC analysis since our one Linux lab is rigged up with CentOS like pseudo cluster and we can just launch jobs from the command line using 1 core and then let the OS take over the whole machine for a while. Though technically the 100Mbps ethernet interconnect is useless for anything higher than what the HPC class handles.

 

Power8 is a bit more squirrelly to get good programs done on. Since you need a minimum of 8 threads per core to launch all cores above the first, and since SMT doesn't scale nicely with vector-based approaches, it's not easy to convert to.

[Dabombinable]

ill be glad to check it out ^^ 

oh fancy pantsies. we get i5 45XXs in classrooms. Though in our lab we have E7v3s

To this day my LGA775 desktop and this laptop are better than anything my old high school still use. (My mum works there-and they still haven't done any upgrades-so even she has a better computer than those found there). Though that school still uses Celeron M380 laptops with 512MB RAM so yeah......

[patrickjp93]

So let's be very clear here... this is what you're saying:

"Any chance we could use this as a chance to literally rewrite Android and inherently destroy compatibility with at least 50% of what's on the app store now.... if not more?"

 

 

lol, pretty much what It hought

 

Android isn't written in Java, and legacy support through the JVM built into the OS is not that difficult. Would people stop spouting off when they have no clue about how this stuff works?

 

Changing JDKs is not going to do anything to existing Java-based apps (but seriously, just write it in C++ anyway), and moving to OpenJDK does not require a single change in the Android OS itself, only which JDK gets packaged with it. The threading model, the scheduler, and everything else remain (almost) exactly the same.

[patrickjp93]

To this day my LGA775 desktop and this laptop are better than anything my old high school still use. (My mum works there-and they still haven't done any upgrades-so even she has a better computer than those found there). Though that school still uses Celeron M380 laptops with 512MB RAM so yeah......

 

Miami University had all of its computer science and engineering computer labs working on Core 2 Duo up until the start of this last semester in August. Now we've moved up to I7 4790 Dell Optiplex machines for all the engineering labs and NUCs for all the others. We also have a Mac lab with Mac Minis attached to Dell displays.

[Jade]

Android isn't written in Java, and legacy support through the JVM built into the OS is not that difficult. Would people stop spouting off when they have no clue about how this stuff works?

 

Changing JDKs is not going to do anything to existing Java-based apps (but seriously, just write it in C++ anyway), and moving to OpenJDK does not require a single change in the Android OS itself, only which JDK gets packaged with it. The threading model, the scheduler, and everything else remain (almost) exactly the same.

You're funny. Java > OpenJDK wasn't the problem. The problem was Misanthrope(? didn't go check who it was)'s suggestion to literally rewrite Android. In C. From top to bottom.

[patrickjp93]

You're funny. Java > OpenJDK wasn't the problem. The problem was Misanthrope(? didn't go check who it was)'s suggestion to literally rewrite Android. In C. From top to bottom.

 

It should be largely rewritten. It's a resource hog compared to iOS or a good Linux Distro. The thread model and all else can still be based on POSIX. And the OS APIs can remain the same even if the undercarriage is changed up. No one's apps have to break.

[Jade]

It should be largely rewritten. It's a resource hog compared to iOS or a good Linux Distro. The thread model and all else can still be based on POSIX. And the OS APIs can remain the same even if the undercarriage is changed up. No one's apps have to break.

Google can barely implement their own design "language" (material design) in their most popular apps (cough Youtube cough)... what makes you think they can and will implement a backwards compatible API for their phone OS, if they ever did rewrite it? And I'm not an Apple fan, not just nitpicking--I use a Nexus 6 as a daily driver.

[patrickjp93]

Google can barely implement their own design "language" (material design) in their most popular apps (cough Youtube cough)... what makes you think they can and will implement a backwards compatible API for their phone OS, if they ever did rewrite it? And I'm not an Apple fan, not just nitpicking--I use a Nexus 6 as a daily driver.

 

Because Google is one of the biggest contributors to Linux and BSD development, wrote its own programming language (GO), developed its own OS around the Linux kernel, and is one of the top software builders in the world. Just because Apple and Microsoft have a headstart doesn't mean Google can't catch up and surpass them.

[Security]

Honestly Oracle is a horrible company and anything moving away from them is a move we should encourage.