|Version 3 (modified by chrisbolt, 8 years ago) (diff)|
SplittedCache in XCache is designed for SMP servers.
By setting xcache.count or xcache.var_count to 2 or more to take the advantage. The recommended value to use is count of your cpu + 1, but you can always try to tune it yourself like other settings.
How it works
On SMP servers, CPUs can access memory in parallel, while the XCache shared memory have to be accessed. By enabling SplittedCache, XCache will split the cache by xcache.count (or xcache.var_count), so both of them can be accessed at the same time.
How is it compares to non-blocking-reader
SplittedCache and nonblocking reader give similar improvement on SMP, but:
- nonblocking read way helps nothing for write
- with SplittedCache it is possible that 2 CPUs need the same cache while the other cache is not needed
(There aren't any tests to prove these words, correct me if I'm wrong)