thanks for sharing this, Julian.
The impact of KPTI is definitely one of the frequently asked questions for us
these days. (Unfortunately it's one of those things where the application
performance impact is difficult to predict, but that shouldn't stop us from trying.)
Not with the IO-500 benchmark suite, but maybe still interesting for some people
on this list could be the BeeGFS performance comparison with and without
client-/server-side KPTI on the public BeeGFS user mailing list, provided by
Guido from Zuse Institute:
Julian Kunkel wrote on 23.01.2018 11:01:
for DKRZ I did some measurements for the performance impact of KPTI on tmpfs:
On our system, there is minimal impact -- I also did a latency
analysis (linked in the post).
I'm curious if other people experienced something else.