The HDS USP-V has an algorithm to prevent the cache from becoming too full. It operates as follows:

 30% - start scanning cache queues for dirty data to destage
 40% - scanning activity accelerates
 50% - destaging becomes a priority, I/O throttled back
 70% - All host I/0 delayed until cache dirty data falls below 50%

To maintain maximum efficiency it is recommended that the cache write pending doesn't exceed 30%

 Front End CHP utilisation < 70%
 Back End DKP utilisation < 50%
 Array Group utilisation < 50% 
 LDEV utilisation < 50%

Recent Changes