Two questions about max heap sizes and available memory in android

Question

I see that the Heap Size is automatically increased as the app needs it, up to whatever the phone's Max Heap Size is. I also see that the Max Heap Size is different depending on the device.

So my first question is, what are the typical Max Heap Sizes on Android devices? I have tested memory allocation on one phone that was able to use a heap over 40mb while another gave out OutOfMemory errors in the 20's mbs. What are the lowest that are in common use and what are the highest that are on common devices? Is there a standard or average?

The second question, and more important one, is how to ensure you are able to use the resources available per device but avoid using too much? I know there are methods such as onLowMemory() but those seem to be only for the entire system memory, not just the heap for your specific application.

Is there a way to detect the max heap size for the device and also detect when the available heap memory is reaching a low point for your application?

For example, if the device only allowed a max heap of 24mb and the app was nearing that limit in allocation, then it could detect and scale back. However, if the device could comfortably handle more, it would be able to take advantage of what is available.

Thanks

Answer 1

Early devices had a per-app cap of 16MB. Later devices increased that to 24MB. Future devices will likely have even more available.

The value is a reflection of the physical memory available on the device and the properties of the display device (because a larger screen capable of displaying more colors will usually require larger bitmaps).

Edit: Additional musings...

I read an article not too long ago that pointed out that garbage-collecting allocators are essentially modeling a machine with infinite memory. You can allocate as much as you want and it'll take care of the details. Android mostly works this way; you keep hard references to the stuff you need, soft/weak references to stuff you might not, and discard references to the stuff you'll never need again. The GC sorts it all out.

In your particular case, you'd use soft references to keep around the things that you don't need to have in memory, but would like to keep if there's enough room.

This starts to fall apart with bitmaps, largely because of some early design decisions that resulted in the "external allocation" mechanism. Further, the soft reference mechanism needs some tuning -- the initial version tended to either keep everything or discard everything.

The Dalvik heap is under active development (see e.g. the notes on Android 2.3 "Gingerbread", which introduces a concurrent GC), so hopefully these issues will be addressed in a future release.

Edit: Update...

The "external allocation" mechanism went away in 4.0 (Ice Cream Sandwich). The pixel data for Bitmaps is now stored on the Dalvik heap, avoiding the earlier annoyances.

Recent devices (e.g. the Nexus 4) cap the heap size at 96MB or more.

A general sense of the app's memory limits can be obtained as the "memory class", from ActivityManager.getMemoryClass(). A more specific value can be had from the java.lang.Runtime function maxMemory().

Answer 2

Here are the "normal" (see below) heap sizes for some specific devices:

• G1: 16MB
• Moto Droid: 24MB
• Nexus One: 32MB
• Viewsonic GTab: 32MB
• Novo7 Paladin: 60MB

I say "normal" because some versions of Android (e.g., CyanogenMod) will allow a user to manually adjust the heap limit. The result can be larger or smaller than the "normal" values.

See this answer for additional information, including how to find out what the heap size actually is programmatically, and also how to distinguish between the absolute heap size limit on the one hand and the heap limit that you should ideally respect, on the other:

Detect application heap size in Android

To detect what your present heap utilization is, you could try using the Runtime class' totalMemory() method. However, I've read reports that different versions/implementations of the Android OS may have different policies regarding whether native memory (from which the backing memory for bitmaps is allocated) is counted against the heap's maximum or not. And, since version 3.0, the native memory is directly taken from the application's own heap.

The iffiness of this calculation makes me think that it is a mistake to monitor your app's usage of memory at runtime, constantly comparing it to the amount available. Also, if you are in the middle of an involved computation, and find that you're running out of memory, it is not always convenient or reasonable to cancel that computation, and it may create a bad experience for your users if you do.

Instead, you might try preemptively defining certain modes, or constraints, upon your app's functional behavior that will ensure that it comes in under whatever your current device's relevant heap limits are (as detected during your app's initialization).

For example, if you have an app that uses a large list of words that must be loaded into memory all at once, then you might constrain your app so that for smaller heap limits only a smaller list of the more common words can be loaded, while for larger heap limits a full list containing many more words can be loaded.

There are also Java programming techniques that allow you to declare certain memory to be reclaimable by the garbage collector on demand, even if it has existing "soft" (rather than hard) references. If you have data that you would like to keep in memory, but which can be re-loaded from non-volatile storage if required (i.e., a cache), then you might consider using soft references to have such memory automatically freed when your app starts bumping against the upper limits of your heap. See this page for info on soft references in Android:

http://developer.android.com/reference/java/lang/ref/SoftReference.html