<   Previous Post  2008 Should be fun
One box fits all -...  Next Post:   >

Comments (2)
  • Add a Comment
  • Edit
  • More Actions v
  • Quarantine this Entry

1 localhost commented Permalink

I think it's a good thing that TPC alert tresholds can be set, but I'm still looking for what settings to use. I know that it depends, but for a DS8000 or DS6000 there must be some good rules to use. I'm specially looking for the (read/write) response time values of the total device.so I'm looking foward to the coming TPC aspects.

2 localhost commented Permalink

As you say, it does depend on what your applications will tolerate. The caution I'd give over using a "total device" setting is that this is an average, so you may have a very poorly responding rank that ends up only skewing the device average by 1ms or so.<div>&nbsp;</div> I guess for your environment it would be a case of sampling the general 'good performance' you see over a period of days / weeks and setting something slightly higher than this - with some degree of buffer. If you start getting alerted too often then increasing the buffer.<div>&nbsp;</div> However as a vague guideline (and I'm no real expert on DS8/6K - but from looking at plenty of SVC statistics from field analysis, for a R5 rank I'd expect no more than 10ms of latency on reads (in general well behaving not overloaded ranks should give much closer to 5ms) (of course 10K vs 15K drives make a difference) and 2-4ms for writes (writes will depend on the actual write bandwidth being sent to the controller and the corresponding amount of cache in the box)<div>&nbsp;</div> If it were me, I'd set 12ms for reads and 5ms for writes as a starting point.