On 6/6/2019 2:51 AM, Sagi Grimberg wrote:
You just doubled the number of resources (mrs+page_vectors) allocated
for a performance optimization (25% in large writes). I'm asking myself
if that is that acceptable? We tend to allocate a lot of those (not to
mention the target side).
We're using same amount of mkey's as before (sig + data + meta mkeys
vs. sig + internal_klm + internal_mtt mkey).
But each mkey has twice of mtt/klm space..
No. It's the same size (I reduced the HCA cap, remember ?)
And we save their invalidations (of the internal mkeys).
Those are not resources.
I'm not sure what is the correct answer here, I'm just wandering if
this
is what we want to do. We have seen people bound by the max_mrs
limitation before, and this is making it worse (at least for the pi
case).
it's not (see above). The limitation of mkeys is mostly with older
HCA's that are not signature capable.
Its also sw mr page_vec/sgl... But that is less of a concern as well I
guess.
