Description of problem: Dictionary is a heavily used data-structure in glusterfs, and knowing its usage patterns will enable us to bring in some enhancements regarding performance. Once easy way would be to utilize the mempool for dict allocations. Version-Release number of selected component (if applicable): upstream
CHANGE: http://review.gluster.com/2851 (core: dict allocations through mem-pool) merged in master by Vijay Bellur (vijay)
Now dictionary is allocated from the mempool. Information about dict usage can be obtained from statedump of the process. Checked with glusterfs-3.3.0qa43. pool-name=glusterfs:data_t hot-count=75 cold-count=16309 padded_sizeof=52 alloc-count=2451087 max-alloc=211 pool-misses=0 max-stdalloc=0 -----=----- pool-name=glusterfs:data_pair_t hot-count=122 cold-count=16262 padded_sizeof=68 alloc-count=2730070 max-alloc=268 pool-misses=0 max-stdalloc=0 -----=----- pool-name=glusterfs:dict_t hot-count=43 cold-count=4053 padded_sizeof=84 alloc-count=537697 max-alloc=73 pool-misses=0 max-stdalloc=0