[DUG] EMBARCADERO MY GRIPE

Thu Jul 30 11:36:08 NZST 2015

Real world performance of code is more complicated than whether it is
compiled native or not, especially when you are talking about a virtual
machine running within or on top of a foreign environment.  Xamarin.Android
code may well be native code, but how efficient that code is comes down to
the original compiler and then also the JIT compiler.

In addition, most Android applications will spend a lot of time invoking
services of the Android platform.itself, so the performance of the code
calling those services may have only a very slight bearing on the overall
performance of the application.  Far more significant is the efficiency of
calls made between the application code and the platform services.  For any
non-platform native approach (Xamarin, FireMonkey) there is necessarily a
"bridge" between the application code and the platform, so any performance
advantage of the native code on one side that bridge has to offset the
overhead of that bridge before it can offer any real advantage.

And with ART, the platform native application code is now also native code
so any advantage there is now - theoretically - entirely lost, leaving only
the overhead of the runtime/platform bridge.

Xamarin's benchmark tests execution efficiency of their generated code in
isolation.  This is a meaningless benchmark since it entirely ignores the
real world impact of the overhead necessarily incurred by having to
constantly bridge from their execution environment to the platform services
on the device.  Maybe Xamarin is faster there too, but if so why not
publish some benchmarks demonstrating that to be the case, rather than
benchmarks which have no relevance ?  Again: Consider the source.

Similarly mobile apps tend to spend a lot of time making network calls.  As
you yourself observed, that can make a big difference in perceived
performance which has nothing what-so-ever to do with the execution
efficiency of the code *making* the network call.

More importantly, all software has bugs - the more software you involve in
a solution, the greater the chance of encountering a bug that will
adversely affect your application and the harder it could be to identify
which component is responsible and obtain a resolution from whichever
vendor is responsible for that particular component in your stack.  e.g.
such as the codegen error in .net framework 4.6 which can introduce bugs
even in existing applications.

Consider the stacks:

Xamarin:     Xamarin / Mono compiler > Xamarin + Mono frameworks > Mono
runtime > Android
Delphi:        Delphi compiler > FireMonkey framework > FireMonkey RTL >
Android
Elements:   Elements compiler > Android

On 30 July 2015 at 11:11, Leigh Wanstead <leigh.wanstead at gmail.com> wrote:

> Hi Jolyon,
>
> The fastest code on android is native code which is compiled by c code.
> Xamarin Android is based on runtime library which I guess is compiled in C
> too. Microsoft's net framework is compile .net code into native code before
> run the byte code on the real device.
>
> Regards
> Leigh
>
> On 30 July 2015 at 11:07, Leigh Wanstead <leigh.wanstead at gmail.com> wrote:
>
>> Hi Jolyon,
>>
>> If network round trip time has little or nothing to do with framework,
>> how you explain that url get different time from different framework? I
>> will assume that they will get similar time spent to get data on all
>> framework.
>>
>> Dalvik platform is slow which is agree by google. Dalvik is slower than
>> Sun's jdk on mobile platform I read somewhere on internet. The
>> consideration for dalvik is not speed, but app size.
>>
>> Regards
>> Leigh
>>
>> On 30 July 2015 at 09:06, Jolyon Smith <jsmith at deltics.co.nz> wrote:
>>
>>> But Leigh, network round trip times have little or nothing to do with
>>> Mono / Dalvik / ART.
>>>
>>> I shall leave the last word on Xamarin to someone else...
>>> http://www.whitneyland.com/2015/07/xamarin-review-2015.html
>>>
>>> I would also recommend reading the earlier post from the same author.
>>>
>>> Worth noting in these round-ups is the point about the lack of community
>>> assistance when it comes to finding Xamarin solutions to common platform
>>> issues (as opposed to the bugs and issues in Xamarin itself).  As mentioned
>>> before, RemObjects Elements avoids this problem due to the fact that the
>>> solutions for Java / Objective-C from the "native" communities for those
>>> platforms, can be applied *directly* in Elements projects in a way that
>>> is not often possible with Xamarin.
>>>
>>> On 29 July 2015 at 15:57, Leigh Wanstead <leigh.wanstead at gmail.com>
>>> wrote:
>>>
>>>> Hi Jolyon,
>>>>
>>>> It seems that we are going through the benchmark way :-)
>>>>
>>>> I tried to run the app in the url you mentioned and it crashed.
>>>>
>>>> How about you look at this url?
>>>> http://magenic.com/Blog/Post/4/Mobile-Development-Platform-Performance
>>>>
>>>> My work is getting data from server which is similar to test 3.
>>>> java version shows 2.369s and xamarin version shows 1.738s in that url.
>>>> That is around half seconds difference.
>>>>
>>>> I sometimes got around less than 70ms round trip time in my own test to
>>>> get data from server in sydney, Australian in north shore, Auckland, nz if
>>>> the server is not busy. That is amazing fast using Xamarin android.
>>>>
>>>> Most customers are in Australia. I guess that they might get around
>>>> 50ms round trip time.
>>>>
>>>> Regards
>>>> Leigh
>>>>
>>>>
>>>> On 29 July 2015 at 14:42, Jolyon Smith <jsmith at deltics.co.nz> wrote:
>>>>
>>>>> ... and if only I had a million dollars I would be rich.
>>>>>
>>>>>
>>>>> As for Xamarin performance, consider the source.  By which I don't
>>>>> mean the code, I mean who is making what claims.
>>>>>
>>>>>
>>>>> http://stackoverflow.com/questions/17134522/does-anyone-have-benchmarks-code-results-comparing-performance-of-android-ap
>>>>>
>>>>> Any advantage is only seen in an Intel Android VM.  On ARM (by far the
>>>>> most prevalent in terms of actual Android hardware), Dalvik beat Xamarin
>>>>> almost every time, until Xamarin.Android 4.7.11.
>>>>>
>>>>> What is odd about this is that these results are from 2013, over a
>>>>> year after Xamarin posted their claims about *astonishingly* superior
>>>>> performance vs Dalvik.  It is interesting that Xamarin do not disclose what
>>>>> environment their benchmarks were run in.  Also interesting that they do
>>>>> not compare themselves to ART which is the more relevant comparison going
>>>>> forward.
>>>>>
>>>>> In any event, I don't think there is any chance that Google will drop
>>>>> ART any time soon (they already dropped Dalvik) in favour of a Mono based
>>>>> implementation of Android.  ;)
>>>>>
>>>>>
>>>>>
>>>>> On 29 July 2015 at 13:51, Leigh Wanstead <leigh.wanstead at gmail.com>
>>>>> wrote:
>>>>>
>>>>>> Hi Jolyon,
>>>>>>
>>>>>> I mentioned to you before in the thread. If google choose to use mono
>>>>>> framework in android, xamarin apk size can reach several kb too. The reason
>>>>>> for me to use Xamarin is the app developed by Xamarin using mono framework
>>>>>> is faster than dalvik before ART time. The load time for the app is not my
>>>>>> main concern. I care about the speed running the app for whole lifecycle.
>>>>>> Here is the url https://blog.xamarin.com/android-in-c-sharp/
>>>>>>
>>>>>> Regards
>>>>>> Leigh
>>>>>>
>>>>>> On 29 July 2015 at 13:40, Jolyon Smith <jsmith at deltics.co.nz> wrote:
>>>>>>
>>>>>>> What a fabulous attitude.  It's thanks to that sort of thinking
>>>>>>> that we now "need" machines with quad core 2.5GHz processors and 8GB of RAM
>>>>>>> just to run frikkin MS Word.
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> NZ Borland Developers Group - Delphi mailing list
>>>>>>> Post: delphi at listserver.123.net.nz
>>>>>>> Admin: http://delphi.org.nz/mailman/listinfo/delphi
>>>>>>> Unsubscribe: send an email to delphi-request at listserver.123.net.nz
>>>>>>> with Subject: unsubscribe
>>>>>>>
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> NZ Borland Developers Group - Delphi mailing list
>>>>>> Post: delphi at listserver.123.net.nz
>>>>>> Admin: http://delphi.org.nz/mailman/listinfo/delphi
>>>>>> Unsubscribe: send an email to delphi-request at listserver.123.net.nz
>>>>>> with Subject: unsubscribe
>>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> NZ Borland Developers Group - Delphi mailing list
>>>>> Post: delphi at listserver.123.net.nz
>>>>> Admin: http://delphi.org.nz/mailman/listinfo/delphi
>>>>> Unsubscribe: send an email to delphi-request at listserver.123.net.nz
>>>>> with Subject: unsubscribe
>>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> NZ Borland Developers Group - Delphi mailing list
>>>> Post: delphi at listserver.123.net.nz
>>>> Admin: http://delphi.org.nz/mailman/listinfo/delphi
>>>> Unsubscribe: send an email to delphi-request at listserver.123.net.nz
>>>> with Subject: unsubscribe
>>>>
>>>
>>>
>>> _______________________________________________
>>> NZ Borland Developers Group - Delphi mailing list
>>> Post: delphi at listserver.123.net.nz
>>> Admin: http://delphi.org.nz/mailman/listinfo/delphi
>>> Unsubscribe: send an email to delphi-request at listserver.123.net.nz with
>>> Subject: unsubscribe
>>>
>>
>>
>
> _______________________________________________
> NZ Borland Developers Group - Delphi mailing list
> Post: delphi at listserver.123.net.nz
> Admin: http://delphi.org.nz/mailman/listinfo/delphi
> Unsubscribe: send an email to delphi-request at listserver.123.net.nz with
> Subject: unsubscribe
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://listserver.123.net.nz/pipermail/delphi/attachments/20150730/1851b7b3/attachment.html 