Cache mechanism
| Project: | JNode Core |
| Component: | Code |
| Category: | task |
| Priority: | normal |
| Assigned: | Fabien D |
| Status: | duplicate |
Jump to:
Description
- Develop a cache system for devices (classes that are instanceof BlockDeviceAPI)
- Study the need to replace the byte[] parameters in BlockDeviceAPI by ByteBuffer or something else
- Study the impact on the classes BlockAlignmentSupport and FSBlockAlignmentSupport
The cache system :
- will be parameterizable (read strategy, write strategy, ...)
- won't be automatically installed on all devices (for example: RAM devices don't need a cache because they are fast enought)
- Login to post comments
byte[] vs. ByteBuffer
I would like very much to have the byte[] parameters in BlockDeviceAPI to be replaced by ByteBuffer's.
Ewout
ByteBuffer: direct or not ?
I am very new to java.nio (and so ByteBuffers) and I am not sure of what is better for performance : direct ByteBuffer or indirect ByteBuffer ?
If there is a difference, I think it can be at assembler/CPU level (CPU access to memory access through pointers).
Do you know if such a difference exists or if it is the same?
However, the ByteBuffer can handle the 2 cases, so there is just one line of code to modify (call to ByteBuffer.allocate(...) or to ByteBuffer.allocateDirect(...)).
Fabien
Re: direct or not
The difference is currently very limited. A direct buffer uses unmovable OS memory instead of java heap memory (for non-direct buffers), but in JNode this is the same.
Anyway for the implementation, there is only a difference on the lowest level, the way data is get out of or into the buffer.
Ewout
#1
This issue is the same as that other issue.