matyix · February 24, 2014 13:26
diff --git a/gistfile1.txt b/gistfile1.txt
 If you are a HBase user a good experience to leave the comfort zone of the JVM heap is to start playing with the new BlockCache implementations in latest HBase versions (made me a sleepless night).

 Older HBase implementations supported a single BlockCache implementation **(LruBlockCache)** but further version have introduced different cache implementations *(a block cache is a single unit of I/O, the smallest amount of data HBase writes or reads back from an HFile)*

 Since the introduction of **ByteBuffers** in Java we had the chance to leverage the operating system's memory manager to move data in and out of memory in a way that's transparent to your program - and this is what exactly the new **SlabCache** and **BucketCache** implementations does, allocates memory outside the JVM heap using direct byte buffers. 

 ``` java java.nio.ByteBuffer class
 public static ByteBuffer allocateDirect(int capacity)
 ```

 The first impression is a faster GC time using the CMS collector. Nevertheless, I think a deeper profiling is required as since the direct buffer leaves outside the JVM heap every data access is a JNI call, so it might work for a few and not for the rest.
	If you are a HBase user a good experience to leave the comfort zone of the JVM heap is to start playing with the new BlockCache implementations in latest HBase versions (made me a sleepless night).

	Older HBase implementations supported a single BlockCache implementation (LruBlockCache) but further version have introduced different cache implementations (a block cache is a single unit of I/O, the smallest amount of data HBase writes or reads back from an HFile)

	Since the introduction of ByteBuffers in Java we had the chance to leverage the operating system's memory manager to move data in and out of memory in a way that's transparent to your program - and this is what exactly the new SlabCache and BucketCache implementations does, allocates memory outside the JVM heap using direct byte buffers.

	``` java java.nio.ByteBuffer class
	public static ByteBuffer allocateDirect(int capacity)
	```

	The first impression is a faster GC time using the CMS collector. Nevertheless, I think a deeper profiling is required as since the direct buffer leaves outside the JVM heap every data access is a JNI call, so it might work for a few and not for the rest.