On 6/28/2011 9:18 AM, Jeff Moyer wrote:

 From the perf report, I would have guessed that the CPU utilization for
the ahci test case would have been lower than the Micron block driver.
Odd, I wonder what I'm missing.  Asai, did you notice if any of the CPUs
was completely pegged during testing with ahci?

Yes, whichever cpu was handling interrupts was pegged in both cases. 
The driver is configured to use a single MSI interrupt.


You're using a NUMA

box, right?  I also wonder what the irq distribution looked like, and
whether rq_affinity is hurting performance for the ahci case.

No, it's not a NUMA box.  Our runs were done with rq_affinity=0.  If you 
think it would be helpful, we can re-run with rq_affinity=1.


   Also,

does the Micron driver do any sort of interrupt coalescing that maybe
the ahci driver isn't doing?

We have a scheme to coalesce multiple completions into a single 
interrupt.  Both drivers were using this feature for these runs.