I think I can see the light
Ok. It seems that I've almost got it done. My eyes hurt so I won't really write much.
Things I want to say:-
First of all, the solution to the memory management: C requires the programmer to manage their memory by themselves - so this is how I should do it. Whenever I open an object for write, it should be the programmer who's responsible for deallocating the memory for that object when they're done with it.
I started reading one of the papers Jim sent, here's an interesting quote from the abstract of the first one:
[Source: Lowering the Overhead of Nonblocking Software Transactional Memory, Marathe et al, 2006]
What this shows is that I beleive it will be really difficult to have a lock-free and garbage collector free STM implementation. But remember to investigate the CAS possibility.
To be honest, I haven't read the papers yet. I will read them tomorrow. You hear me Fuad, READ THE PAPERS TOMORROW! :)
Things I want to say:-
First of all, the solution to the memory management: C requires the programmer to manage their memory by themselves - so this is how I should do it. Whenever I open an object for write, it should be the programmer who's responsible for deallocating the memory for that object when they're done with it.
I started reading one of the papers Jim sent, here's an interesting quote from the abstract of the first one:
Most either employ locks in the underlying implementation or depend on thread-safe general-purpose garbage collection to collect stale data and metadata.
[Source: Lowering the Overhead of Nonblocking Software Transactional Memory, Marathe et al, 2006]
What this shows is that I beleive it will be really difficult to have a lock-free and garbage collector free STM implementation. But remember to investigate the CAS possibility.
To be honest, I haven't read the papers yet. I will read them tomorrow. You hear me Fuad, READ THE PAPERS TOMORROW! :)
posted by Fuad at
3:23 PM
0 Comments:
Post a Comment
<< Home