mazurkin / opl

Off-Heap Primitives Library

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

OPL - Off-heap Primitives Library

OPL (Off-heap Primitives Library) implements a set of structures and algorithms which operate in off-heap memory.

  • Structures can have size more than 2GB
  • all addressing by long not int
  • Fast operation without bound check (safety bound check is also available as an option)

JVM

org.opl.platform.Jvm wraps all calls to sun.misc.Unsafe make using these operations more comfortable without producing any compiler and import warning.

Jvm.putLong(address, value);

Allocators

Allocator allows to allocate, reallocate and free memory blocks.

Allocator allocator = new SystemAllocator();
try {    
    long address = allocator.allocate(1024);
    try {        
        Jvm.putLong(address + 8, 0x1111_2222_3333_4444L);
        long v = Jvm.getLong(address + 8);
    } finally {
        allocator.free(address);
    }
} finally { 
    allocator.close();
}

MemoryBlockFactory and MemoryBlock

MemoryBlock describes memory block as a pair (address, size) and has some address arithmetic methods.

MemoryBlockFactory is a wrapper around Allocator to allocate, reallocate and free memory block instances.

MemoryBlockFactory factory = 
    new MemoryBlockFactory(new SystemAllocator(), true, false);
    
try {
    MemoryBlock block = factory.allocate(1024);
    try {        
        Jvm.putLong(block.address(8), 0x1111_2222_3333_4444L);
        long v = Jvm.getLong(block.adress(8));
    } finally {
        factory.free(block);
    }
} finally {
    factory.close();
}    

Structures

OplBitSet

Off-heap bit set

OplBitSet bitset = new OplBitSetImpl(memBlkFactory, 3 * Mem.GB);
try {
    bitset.set(10000);
    bitset.toggle(10001);
} finally {
    bitset.close(); // free memory
}       

OplBloomFilter

Off-heap Bloom filter

OplBloomFilter filter = new OplBloomFilterImpl(memBlkFactory,
try {
    filter.put(0x0123);
    Assert.assertTrue(filter.contains(0x0123)); // always true
    Assert.assertFalse(filter.contains(0x0567)); // actually may fail because the structure is probabilistic 
} finally {
    filter.close(); // free memory
}

Safety

Programming with off-heap is not safe as you can crash the JVM instead of getting well-structured exception.

So on development/test phase use safety options.

MemoryBlockFactory

MemoryBlockFactory can allocate safe block which check indexing on operations.

# Create a factory that produces "checking" blocks
boolean safe = true;

MemoryBlockFactory factory = 
    new MemoryBlockFactory(new SystemAllocator(), true, safe);

MemoryBlock block = factory.allocate(1024);

// IndexOfBoundException - not SIGSERV  
block.address(1025); 

RegistryAllocatorProxy

RegistryAllocatorProxy is a wrapper real allocator that helps to track all allocated block addresses

Allocator allocator = new RegistryAllocatorProxy(
    new SystemAllocator(), true);
    
// We won't free this address
long address = allocate.allocate(1024);
    
// Release some invalid address - AllocatorException is raised
allocator.free(0x1234);      

// will get IllegalStateException as one block has not been released
allocator.close();

DefensiveAllocatorProxy

DefensiveAllocatorProxy allocates safety areas around the real block so it can detect dangerous operation made outside a block.

Allocator allocator = new DefensiveAllocatorProxy(  
    new SystemAllocator(), true);
    
// We will make an illegal write
long address = allocate.allocate(1024);

// Write outside the allocated block
Jvm.putInt(address + 1024, 0x1234);

// will get AllocatorException as the block is corrupted
allocate.free(address);

You can combine both proxies:

 Allocator allocator = new DefensiveAllocatorProxy( 
    new RegistryAllocatorProxy(new SystemAllocator(), true), 
    true);      

About

Off-Heap Primitives Library

License:Apache License 2.0


Languages

Language:Java 100.0%