On Thu, Feb 05, 2015 at 03:17:19PM +0100, Kevin Wolf wrote: > > By never allowing the hit count to go down to zero, we make sure > > that all unused entries are chosen first before a valid one is > > discarded. > > But does this actually improve a lot? cache_hits is only 0 for the > first few accesses and it never becomes 0 again after that. The > result might be that soon all the entries have cache_hits == 1, and > we get the same problem as you're describing - only the first few > entries will be reused.
I targeted the specific case where the size of the L2 cache is set so that it's big enough for the whole disk. I have a setup with a 16GB disk image and a 2MB L2 cache (that's 32 entries). If I do random rw, after two minutes I get an average of 3K IOPS, and more than half (19) of the cache entries are still unused. With the patch the cache fills up immediately and I get sustained ~20K IOPS from the very beginning. Berto
