[PATCH v7] NVMe: conversion to blk-mq

[Jens Axboe]

On 06/13/2014 01:22 PM, Keith Busch wrote:
> One performance oddity we observe is that servicing the interrupt on the
> thread sibling of the core that submitted the I/O is the worst performing
> cpu you can chose; it's actually better to use a different core on the
> same node. At least that's true as long as you're not utilizing the cpus
> for other work, so YMMV.

This doesn't match what I see here. Just ran some test cases - both
sync, and higher QD. For sync performance, core or thread sibling is the
best choice, other CPUs next. That is pretty logical.

For a more loaded run, thread sibling ends up being a better choice than
core, since core runs out of steam (255K vs 275K here). And thread
sibling is still a marginally better choice than some other core on the
same node.

Which pretty much matches my expectations of what the best mappings
would be.