Not sure when you last checked LtU, but I just posted a paper on a competitive refcounting GC based on Immix:

http://lambda-the-ultimate.org/node/4825

They conclude that the heap organization and resulting cache misses account for at least 10% throughput degradation of refcounting GCs compared to tracing.

Sandro

On 11/10/2013 12:26 PM, Jonathan S. Shapiro wrote:
Yes. The Berger paper is very helpful. I also found some measurements of L1 cache misses (which are a good approximation to overall overhead) in a size-fitting allocator.

The original impetus for my question was the astonishing overhead of reference counting. But my surprise was misplaced. The right question is the /relative/ overhead of RC when compared to malloc/free. Surprisingly, they are very close.


Jonathan


_______________________________________________
bitc-dev mailing list
[email protected]
http://www.coyotos.org/mailman/listinfo/bitc-dev


Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

_______________________________________________
bitc-dev mailing list
[email protected]
http://www.coyotos.org/mailman/listinfo/bitc-dev

Reply via email to