On 06/20/2013 05:23 PM, Shaun Thomas wrote:
> On 06/20/2013 03:32 PM, Josh Berkus wrote:
>
>> Did you compare setting RPC to 1.0 vs. setting it to 1.1, or something
>> else just slightly higher than SPC?
>
> Yes, actually. My favored setting when we were on 8.3 was 1.5. But
> something with the planner changed pretty drastically when we went to
> 9.1, and we were getting some really bad query plans unless we
> *strongly* suggested RPC was cheap. I was afraid I'd have to go lower,
> but 1 seemed to do the trick.
>
That would be perverse, surely, but on Fusion-IO RPC = SPC seems to make
sense unless you assume that cache misses will be higher for random
reads than for sequential reads.
cheers
andrew