[Crossposted to comp.lang.java.programmer, since more people lurk there]
>>> People, how can I know how big an object is? I mean, I have an object
>>> the
[quoted text clipped - 5 lines]
>>>
>>> How do I do such thing?
I wrote:
>> Are you sure it's the object's size you need, and not, for
>> example, the
>> size of the buffer within that object?
>>
>> How about whenever you collect a unit of data, you increment a
>> counter?
OP responds:
> Actually, I'm not! The thing is, I've already seen the JVM misbehave when
> it uses a lot of memmory. Lets say, for example, If I have 750 MB of RAM
[quoted text clipped - 10 lines]
> counter is not the best way sisnce the size of the headers may differ from
> message to message.
Not sure the best way to solve this one, which is why I'm crossposting
it to c.l.j.programmer, so others can contribute.
I've read about one technique but never tried it: Allocate a big chunk
of memory at the start of your program, do your memory operation which might
result in an OutOfMemoryException (OOME), catch it, release the big chunk of
memory that you got, so you now have some wiggle room to work with,
serialize the buffer to disk, re-allocate another big chunk of memory (and
assert that it doesn't fail, since you're supposed to have a lot of free
memory after the serialization), and repeat.
- Oliver
Robert Klemme - 29 May 2006 16:09 GMT
> [Crossposted to comp.lang.java.programmer, since more people lurk there]
>
[quoted text clipped - 45 lines]
> chunk of memory (and assert that it doesn't fail, since you're supposed
> to have a lot of free memory after the serialization), and repeat.
In this scenario I'd just count downloaded bytes (I guess this is what
you meant by "counter" above). As easy as that. With a limit of 250MB
some bytes overhead for reference to an object's class etc. are negligible.
And I think you also correctly pointed out that "the size of an object"
is a too fuzzy term to deal with. Strictly speaking, the size of an
object is 8 bytes + 4 * size of an object reference + size of any POD
data members (roughly, I'm not 100% sure about the figures). But then
again, for a String you would certainly also want to count the size of
the char array that holds the actual string data etc. Matters become
even more complicated when two strings share the same array, e.g. you
use this constructor: public String(String original). Does the char
array count for each of them? Or is half the length counted for each of
them? This becomes even more complex with collections and user defined
classes...
Kind regards
robert
Oliver Wong - 29 May 2006 16:44 GMT
>> "Scirious" <scirious@scirious.com> wrote in message
>>> So, for example, I'm writing an e-mail client and I have
>>> an object to store the headers I download from the server. I may want to
>>> specify that, after downloading 250 MB of headers, I want them to be
>>> recorded on disc so the object can give place to a new one and be
>>> collected
>>> from memory.
[...]
> In this scenario I'd just count downloaded bytes (I guess this is what you
> meant by "counter" above). As easy as that. With a limit of 250MB some
> bytes overhead for reference to an object's class etc. are negligible.
I think it's a bit trickier than that. Let's say the downloaded bytes
are 0x48, 0x65, 0x6C, 0x6C, 0x6F. I.e. the string "Hello" if decoded using
ASCII. If the program then takes these 5 bytes and stores them as a String,
the JVM is likely to use UTF-16 internally, which will result in 10 bytes
being used (2 for each character).
If the downloaded bytes UTF-8 instead of ASCII, and some of the
characters were above codepoint 256, then the calculation is not simply a
matter of doubling the downloaded bytes. Some characters will have taken 1
byte to download, 2 bytes to store; others will have taken 2 bytes to
download, 2 bytes to store. Still others, 3 bytes to download, 2 bytes to
store; 3 bytes to download, 3 bytes to store, and so on with various other
combinations.
- Oliver
Robert Klemme - 30 May 2006 08:22 GMT
>>> "Scirious" <scirious@scirious.com> wrote in message
>
[quoted text clipped - 24 lines]
> 2 bytes to store; 3 bytes to download, 3 bytes to store, and so on with
> various other combinations.
Maybe I wasn't clear enough: I meant, ignore real sizes and just use the
bytecount as limit - alternatively, if the data is converted to chars
then use the char count as limit. I wouldn't bother to think about
actual sizes in bytes - the important part for me would be that there is
a *limit* to the size.
Kind regards
robert