Hi, Xiaofeng, Sorry for late replay, I just have moved to a new renting room recent days, the network is just connected.
I think I have finished first 2 parts of this project. DONE: In step 0, I find the cause of problem of barrier generating, xiaoming helps me on this issue and give a solution. In step 1, I implemented 2 version of barrier for get() method, one is simply using a JNI call to invoke a GC method. Another version is a vmmagic version, my 'get() barrier' implementation just follows the approach of implementing write barrier. In step 2, I modified the concurrent marker to support weak reference processing, markers threads will put the reference object to a list according to its type. when the concurrent marking finishes, I use the last marker thread to process soft/weak reference lists, concurrently (Because this process do not impact the object referencing graph). Now, I am thinking of a improving, need we make the weak/soft reference processing parallel? TODO: In step 3, I will implement a STW phase for finalizable objects and phantom reference processing. In this phase, gc thread will trace from the dead objects in the list to resurrect all reachable objects from them. this phase will change the object referencing graph, so it should be dealt in a STW manner. I am starting writing the formal mid-term status report today, is there anything I should pay more attention to in this report? Thank you! Simon Thanks 2009/7/2 Xiao-Feng Li <xiaofeng...@gmail.com> > Simon, what's your progress? > > Thanks, > xiaofeng > > -- > http://people.apache.org/~xli > -- >From : simon.z...@ppi, Fudan University
