Why *just* a ring buffer?
Posted Aug 3, 2006 16:34 UTC (Thu) by cventers
In reply to: Why *just* a ring buffer?
Parent article: Toward a kernel events interface
The vector you refer to would likely fall apart as you add more and more
CPUs. If you have several threads doing I/O on several processors, each
CPU would be busy doing atomic word updates to the same cache line, which
would get flushed /constantly/. The greater the data density, the more it
hurts, and a *bitmap* is as dense as you can get.
Keep in mind that I haven't actually measured the effects. It is of
course very difficult at times to avoid caching issues (since even a
simple lock suffers from this problem).
to post comments)