I am currently developing an android application and I am using my own device (nexus 4) to debug it. I seem to have troubles regarding the heap size allocated by the Dalvik VM and I hope you can help me on this one.
My application is processing a relatively important amount of data (~3MB) from a file and use it to generate java objects which go around 10MB in total. This should not be a problem for a device which has 2GB of RAM and a heap size limit of 512MB ( if I am not mistaken).
However, the debugging tool tells me that the heap size allocated to my application is roughly 10MB, leading to an outrageous amount of garbage collection (each GC frees ~500KB). I went through already a lot of pages on SO and other forums about Dalvik heap size model, etc and I found the largeHeap="true" (API level 13+) attribute to use in this case, however, it did not improve anything and I am still stuck with 10MB to use when the device has an initial memory of 2GB. I understand that the OS, etc ,etc take some memory itself, that Dalvik is restricting applications as well, but why can't I reach at least the 512MB?
So, did I misunderstood something? Did I miss something? The problem seems rather simple, but I don't get it.
Your help will be greatly appreciated!
PS: Please don't suggest to switch to the NDK, it is not suitable in my case.
If you're not getting OutOfMemoryError, altering the heap size won't affect the behavior.
Dalvik's managed heap resists growth in an attempt to minimize the app's memory footprint, and unfortunately will sometimes get "hung up" on a particular size. It's usually possible to kluge around this by allocating a large object (say, a byte[1000000]) and immediately dropping the reference to it.
FWIW, the best way to guarantee the reference is dropped is to allocate it in a method that returns immediately. Don't just null out a reference. The Dalvik GC is not live-precise, so it won't ignore "dead" references lingering in registers.
(All of the above is for Android 4.4 and earlier. Hopefully the situation will improve in future releases.)