Steffen Klassert <steffen.klass...@secunet.com> wrote: > On Tue, Jun 19, 2018 at 09:53:49AM +0200, Florian Westphal wrote: > > Li RongQing <lirongq...@baidu.com> wrote: > > > The default NR_CPUS can be very large, but actual possible nr_cpu_ids > > > usually is very small. For some x86 distribution, the NR_CPUS is 8192 > > > and nr_cpu_ids is 4, so replace NR_CPU to save some memory > > > > Steffen, > > > > I will soon submit a patch to remove the percpu cache; removal > > improved performance for at least one user (and by quite a sizeable > > amount). > > > > Would you consider such removal for ipsec or ipsec-next? > > I think this removel would better fit to ipsec-next.
Agree, it slows things down further for me in my tests. Problem is that I get quite good re-use of pcpu cache due to unidirectional flows and only one tunnel. I suspect that even with tunnel the removal is a win in practice though, netperf is quite artifical, so I rather trust Kristians results (real world) than my own. > considered to apply it to ipsec-next. If you plan > to remove it, I'll wait for that. I'll submit once net-next opens.