Customer states his batch job, which normally runs for 12hrs 30mins, is running 2hrs 15mins longer on the VMAX20K.

The latest batch job ran for 14hrs 15mins. I collected a number of performance statistics, some are listed below. The host has a single 1TB, sixteen member thin meta, under FAST VP control with a 100,100,100 policy. The device has 689GB in the FC tier and 187GB in the SATA tier. I've never seen any allocated space move up to the FLASH tier. The host is zoned to 4 FAs and PowerPath is configured.

Reads

     IOPs          11.8 to 1162.9     average 239.2

     MB/s          .1 to 5.1               average 1.1

     KBs             .8 to 43               average 6.7

     RT(ms)        .3 to 5.6             average 1.3

     Hit %            74.9 to 100        average 92

Writes

     IOPs          1.3 to 832.3          average 141

     MB/s          .1 to 4.9                average .75
 

     KBs          .9 to 36.8               average 9.6

     RT(ms)     .4 to 3.6                average .9

     Hit%          98.4 to 100          average 96.4

Prefetch Tracks/sec     .1 to 70.1     average 17

Prefetch Tracks/Used     0

Near as I can see the entire device lives in cache, and response times are excellent. The back-end doesn't appear to be over-loaded, although I'm not sure what penalty, if any, a prefetching brings if no prefetched tracks are used. The front-end does have spikes where it reaches 70% busy , and over, but the low response times and high cache hits I would think trump how busy the FAs are. Outside of the fact the IOs are small, and the reads are 60% random, I'm having a hard time understanding why the perceived performance is so poor.