Good Fowler article on ORM

[Jan Burse]

I don't think I get your last paragraph here. What pain? What additional
classes? Why?

The discussion was about some custom HashMap, call this
class XXX. If one does use inline loops such as:

for (int i = 0; i<table.length; i++) {
Entry e = table;
while (e!=null) {
/* do something */
}
}

The one does not need to defined an interator for the
custom HashMap, call this iterator class YYY.

These iterators which implement the interface Iterator
(or Enumerator) are usually not much visible Java collection
classes, since they are realized as inner classes.

But they add to the LOCs and the number of classes:

With Iterator Without Iterator
XXX N + K LOCs N LOCs
YYY M LOCs
Total N + K + M N

K is for the iterator factories inside the collection
class. Pain = N + K + M - N = K + M.

Since nobody would use class YYY anyway, since XXX
is anyway internal to the whole multi-indexing
framework and not visible by the client of the framework,
its not only a pain but also not a necessity.

I'm willing to lay odds that for your use cases
the difference made by accessors and mutators is
not the low-hanging fruit.

For gambling you can go to Las Vegas, or if you
are up to something more serious, you could educate
yourself.

First of all originally Android didn't have a JIT.
See for example:

There might still be devices around without a JIT.

Second for my setter/getter claim you might want
to read this page:
http://developer.android.com/guide/practices/design/performance.html
There is a section about Avoid Internal Getters/Setters.

I copy the essential part for your eyes here:

"In native languages like C++ it's common practice to use getters (e.g.
i = getCount()) instead of accessing the field directly (i = mCount).
This is an excellent habit for C++, because the compiler can usually
inline the access, and if you need to restrict or debug field access you
can add the code at any time.

On Android, this is a bad idea. Virtual method calls are expensive, much
more so than instance field lookups. It's reasonable to follow common
object-oriented programming practices and have getters and setters in
the public interface, but within a class you should always access fields
directly.

Without a JIT, direct field access is about 3x faster than invoking a
trivial getter. With the JIT (where direct field access is as cheap as
accessing a local), direct field access is about 7x faster than invoking
a trivial getter. This is true in Froyo, but will improve in the future
when the JIT inlines getter methods."

Bye

 

[Jan Burse]

Without a JIT, direct field access is about 3x faster than invoking a
trivial getter. With the JIT (where direct field access is as cheap as
accessing a local), direct field access is about 7x faster than invoking
a trivial getter. This is true in Froyo, but will improve in the future
when the JIT inlines getter methods."

Maybe an updated happened, but I didn't find a notice
about it. The last notice I found was:

http://android-developers.blogspot.com/2010/05/dalvik-jit.html
Which is already 2 years old!

The above blog post refers to Android 2.2,
which is Froyo.

Maybe somehow has found something newer?

Bye

 

[Lew]

Maybe an updated happened, but I didn't find a notice
about it. The last notice I found was:

http://android-developers.blogspot.com/2010/05/dalvik-jit.html
Which is already 2 years old!

The above blog post refers to Android 2.2,
which is Froyo.

Maybe somehow has found something newer?

And how much difference does that make to the application? 1%? 17%? 97%?

"3x faster" and "7x faster" at what level? Just the raw call to get the
attribute? In what usage scenario? At what load on the device?

The effect that has depends on how much those accessors are called. And what
is the cost to the application maintainability?

For a cost-benefit analysis you need to identify both the costs and the
benefits, and accurately, and relevantly.

 

[Lew]

I don't think I get your last paragraph here. What pain? What additional
classes? Why?

The discussion was about some custom HashMap, call this
class XXX. If one does use inline loops such as:

for (int i = 0; i<table.length; i++) {
Entry e = table;
while (e!=null) {
/* do something */
}
}

The one does not need to defined an interator for the
custom HashMap, call this iterator class YYY.

If you have a custom 'HashMap', are you extending 'HashMap'?

You would have to avoid that if you are seeking to micromanage the class load.

These iterators which implement the interface Iterator
(or Enumerator) are usually not much visible Java collection
classes, since they are realized as inner classes.

But they add to the LOCs and the number of classes:

With Iterator Without Iterator
XXX N + K LOCs N LOCs
YYY M LOCs
Total N + K + M N

K is for the iterator factories inside the collection
class. Pain = N + K + M - N = K + M.

Since nobody would use class YYY anyway, since XXX
is anyway internal to the whole multi-indexing
framework and not visible by the client of the framework,
its not only a pain but also not a necessity.

I am not convinced that you are discussing real pain.

You seem to be claiming a reduction in class load, which if true might be
useful if it's significant. How much are you saving in class load?

For gambling you can go to Las Vegas, or if you
are up to something more serious, you could educate
yourself.

Oh, thank you for that, Mr. Snarky One.

Your points don't address mine, however.

You have yet to quote the statistics I requested.

What is the difference in *your* application?

I haven't lost my bet yet.

 

[Robert Klemme]

Thank you for the link Arved!

Now if all programmers who deal with object-to-relational mapping would
only read it.

.... and learn about their relational database. Very true.

Kind regards

robert

 

[Jan Burse]

You have yet to quote the statistics I requested.

I didn't do some measurements, should be
clear from my response to markspace, isn't it?

Bye

 

[Lew]

I didn't do some measurements, should be
clear from my response to markspace, isn't it?

I responded to that post. The only facts you presented in evidence were that
your Android times were slower by some unspecified margin than the desktop
times, and that the Android platform was inherently physically slower than the
desktop.

I suggest that the platform differences, to which I alluded in some detail in
that response, account for the performance differences you observe, or could.
You need measurements that distinguish the effects of the platform from other
effects to know for sure.

At no point have you provided evidence here speaking to what in the
environments could account for the difference, or apportioning impact to known
factors such as the CPU speed and architecture differences.

You probably don't actually know what causes what proportion of the observed
(and undisclosed) difference. It is likely that the largest performance
differences are due to the sorts of factors I mentioned in my other response.

Why do I say "likely"? Because my own observations of Android applications and
my knowledge of the platform and experience with the balance of attribute
access via methods vs. other factors such as I/O, UI manipulation, thread
foobars and the like, I have seen more performance impact from the platform
effects than your suggested micro-optimization.

But I don't know for sure. I haven't got your numbers. It is possible that you
have evidence heretofore unshared that makes getters and setters the
low-hanging fruit after all. So, naturally, I request that you share that
evidence, at least in summary.

Whatever your numbers do show, it makes sense to work first on the matters
that make the biggest difference. If it's UI contention, work on that. If it
really is class load, work on that. If it's logger crud, work on that. What
you might wish to avoid is laboring intensely over an optimization that saves
one part in a thousand of time, causes much more maintenance work due to
workarounds and concomitant complexity induced by the fooferol, and is
obsoleted in a year by platform advances.

YMMV. But you might wish to educate yourself about where your actual problems lie.

 

[Fredrik Jonson]

If you don't use a device but an emulator, you also observe slowdown,
since the Dalvik then runs inside the emulator (sic!).

From my perspectiv having the emulator run dalvik exactly like it would have on
on real devices is a good thing. I want device emulators being as close to
real world devices as possible, if not bug for bug compatible, then at least
having the same memory management pattern as a real device. And I'd definitively
trade some performance for more behaviour similarity.

You just need to learn to go get some coffee or stretch your legs while the
emulator boots.

 

[jb]

Hi Lew,

But I don't know for sure.

You are playing a very cheap game here. Turning a qualitative
discussion into
a quantitative discussion. And putting words into other peoples mouth.

Assume I have stated that something exists, this is like saying x>0.
Now you go on pondering on the degree of the existence, in that you
blaming me for not displaying a margin m, effective value e and
verification e>m.

Who are you to change the intent of my communication? If you are
interested in m, e than do your own research. I gave you the evidence
for >0 by reference to the Android. I wish that you do not anymore
disturbe my post in the future in this way.

Or for short:
Suck my dick, shut up and **** off.

 

[David Lamb]

You just need to learn to go get some coffee or stretch your legs while the
emulator boots.

Stretch your legs by walking to Columbia to pick the coffee beans.

 

[Arved Sandstrom]

What he says is certainly good, but I am not sure
that it is all new.

No, it's not new. Fowler's a respected commentator, though, and it's
good to get his fresh take on a topic. I don't always agree with him but
I think he's way more often right than he's wrong, and I've rarely read
anything by him that I'd dismiss out of hand.

Using an ORM does make all problem disappear. But that
is just a special case of: using tool X does not make
all problem disappear. But making some problems
disappear is still good.

Arne

AHS

 

[Lew]

Or for short:
[expletives deleted].

That was uncalled-for. I assure you the information and ideas I presented are
both well grounded and intended to serve you. Why such a psychotic reaction to
a technical discussion?

It's a rhetorical question. I'm done with you. Buh-bye.

 

[Arne Vajhøj]

You are playing a very cheap game here. Turning a qualitative
discussion into
a quantitative discussion. And putting words into other peoples mouth.

You wrote:

#But I did not yet use it. Was just benchmarking my App and
#saw that it runs much slower on Android. But that has also
#to do with that I was using a tablet with something of 1GHz
#ARM and was comparing against a 3.4GHz 64-bit Intel.

Claiming that that qualitative a Java app is slower on X
than on Y can when X is in general slower than Y can be
abbreviated to "". It provides no real information.

Arne