Adam Olsen wrote: > I'm now working on an approach that writes out refcounts in batches to > reduce contention. The initial cost is much higher, but it scales > better too. I've currently got it to just under 50% cost, meaning two > threads is a slight net gain.
http://www.research.ibm.com/people/d/dfb/publications.html Look at the various papers on 'Recycler'. The way it works is that for each thread, there is an addref buffer and a decref buffer. The buffers are arrays of pointers. Each time a reference is addref'd, its appended to the addref buffer, likewise for decref. When a buffer gets full, it is added to a queue and then a new buffer is allocated. There is a background thread that actually applies the refcounts from the buffers and frees the objects. Since this background thread is the only thread that ever touches the actual refcount field of the object, there's no need for locking. -- Talin _______________________________________________ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
