Hash based persistence - how it works

Hi Piotr,

 

what you are describing is the CARP option (Cache Array Routing Protocol).

 

Yes, hash based score will be calculated for each pool member. Highest one wins.

 

If it goes down, second best will be selected and so on.

 

With CARP no persistence table will be used. So it scales pretty well if you need to persist i.e. on millions of keys.

 

But it takes more CPU cycles to calculate.

 

Traditional hash just calculates a modulus. A change in number of pool members will cause complete redistribution.

 

CARP based method will result in very little redistribution if a pool members are added or removed.

 

Thanks, Stephan