We are using large (20GB), very active (& gt; 92% hit rate) memory cluster and See more than 10% waste space.
The removal rate is extremely low (0.1 / second), and the retrieved rate is flat - 0.
There is reason to worry, and anything like this can be done - On time - to make sure that it is not currently eating a much more useful place? I have recently done an analysis of the space usage of our assembled cluster, which is almost twice the size of that size.
/ P>
We found that there was usually "waste" location in the memcachad In it, the empty slabs were vacant in the empty slab which were not only empty.
This is affected by a side on the fact that the allocation in the slab-memacet is & lt; 1.4.11 is stable and for a fixed slab size the memory is ejected (only once allotted) from the same size slab.
A feature of slab rialocation was introduced as 1.4.11, which is believed to be used on the required slab class to produce LRU objects from other slab classes. Allows to be removed.
I'm not sure whether the removal rate was such,
However, given the low expulsion rate, I'm more than I'm experiencing performance problems Will not be worried.
Hope that helps.
Comments
Post a Comment