Data storage on files powered by NVMe
Project architecture
Currently, the project features a reference allocator storing data on files: MemoryAllocatorOnFile.
Though implemented in a straight-forward way, it still has several underlying chunks depending
on the size of inner blocks it allocates.
The interface MemoryAllocator is the specification for any allocator tested in this project.
This allocator is used to implement Chunks, that represent an array of raw data. ActiveViam having
goals to get the best of the machine performance, Chunks cannot have generic methods, not to pay
the cost of boxing/unboxing. That's the main reason for the existence of Chunk. It provides
read and write methods adapted for primitive types. In this scholar project, we only define methods
for int and double.
The project already ships two sets of implementations:
- the series of
HeapXyzChunk, acting as a reference, using an underlying Java array - the series of
FileXyzChunk, that makes use of theMemoryAllocatorOnFile.
All Chunk implementations are succinctly tested. Those tests can be used as references for testing new implementations of Chunks or Allocators.
Development environment
This project is written in Java 15, though it does not use many of the latest features so far.
It uses Maven to manage its dependencies.
Compiling the project
mvn compileBuild print compilation warnings due to the usage of sun.misc.Unsafe. This is indeed an internal
API but it has been accepted by many vendors. Project Panama
aims at providing safer alternatives, but there are not ready yet.
Running unit tests
mvn test