This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Garbage Collection


I went to a public talk by Jim Miller from MSFT last week. He ran the C#
runtime project for a while. A few interesting points:

1. Their garbage collection guy worked on GC for 8 years straight, full time.
Getting it right is tough.
2. They use an adaptive model where the GC watches the program and then picks
a strategy based on the behaviour.
3. The objective is to keep a working set in level 2 cache.
4. After all that, they are considering a reference counting model to deal
with short lived objects.
5. It's all a lot easier when you are targetting one CPU model. The code knows
all the cache sizes etc.

They have released the code for their run time, including the C# compiler, but
he said the GC and JIT are 'cut down' versions. Supposedly looking at the code
would not taint the reader (for contributing to 'open source' projects), but I
would advise care in this regard.

All in all, not that different to the discussion that went on here.

Tim Josling


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]