Is there a way of estimating (roughly) in memory object size from Serialized object size in Java
-
3[More or less similar here] (stackoverflow.com/questions/3983360/…)sudmong– sudmong2011-08-22 11:09:18 +00:00Commented Aug 22, 2011 at 11:09
-
possible duplicate of Estimate serialization size of objects?Ciro Santilli OurBigBook.com– Ciro Santilli OurBigBook.com2015-05-02 20:08:18 +00:00Commented May 2, 2015 at 20:08
Add a comment
|
2 Answers
The size in memory will be usually between half and double the serializable size. The most extreme example might be the Byte which is more than 80 bytes Serialized can be 16 bytes in memory.
You can use a profiler to tell you how much memory an object uses. Another way is to use a tool based on Instrumentation.getObjectSize(object)
You might find this interesting Getting the size of an Object
14 Comments
Clark Bao
interesting,why bytes is so different. and it is five times larger compared to in memory. Do you have any reference link? +1
Peter Lawrey
Serialization uses a standard format with a header, classes encoded, and their parents. Instead of trying to read a guide on the internal representation of Java Serialization, which I don't belive is an open standard you would learn a lot more by trying it.
Peter Lawrey
To do that you need to how you want to encode each value/field, in which case you should be able to calculate how larger the encoded value will be. This doesn't give you any idea how large is the header, or the impact of the memory alignment of allocation.
Peter Lawrey
A byte[] uses about 12 bytes plus the
length. Most JVMs are 8-byte aligned so you need to round up to the next 8 bytes.Peter Lawrey
@krumpelstiltskin A simple example is a String which uses UTF-8 encoding when serialized. A long ASCII chars string will be 1 byte per char (plus a fixed size header), however in memory it will be one char (2 bytes) per char (unless compressing strings are used)
|
A very nice Tool for this challenge: https://github.com/jbellis/jamm
From the readme.txt:
MemoryMeter is as accurate as java.lang.instrument.Instrumentation.getObjectSize, which only claims to provide "approximate" results, but in practice seems to work as expected.
MemoryMeter uses reflection to crawl the object graph for measureDeep. Reflection is slow: measuring a one-million object Cassandra Memtable (that is, 1 million children from MemoryMeter.countChildren) took about 5 seconds wall clock time.
