4 Jan
2018
4 Jan
'18
9:26 a.m.
On 16/12/17 06:26 +0100, Fabio M. Di Nitto wrote:
On 12/15/2017 9:22 PM, Jan Pokorný wrote:
In the light of the recent KAISER/KPTI Intel et al. affair, it
might also be very interesting to see the actual impact on Kronosnet
on Linux.
For instance[1]:
recently, I've come across this test
comparing OpenVPN vs. SSH
tunnelling:
http://blog.backslasher.net/ssh-openvpn-tunneling.html
so natutally, I guess it would be interesting to perform similar
test cases in a Kronosnet - OpenVPN duel so that the expectations
are set for "VPN on steroids" and perhaps the limitations are
identified if the results are too far from "on par".
(Just saying, not doing, I wouldn't go too far in tweaking
kronosnet for the best results possible, I'm afraid).
That´s an interesting idea to expand in a spreadsheet or something.
Both implementations have pro and cons for sure and some of them are
fairly well known already.
It will make sense to have this done once we rewrite kronosnetd and
there is more than one user (corosync) for knet. Right now, with only
one use case it´s somewhat irrelevant. knet can´t yet compete on the
generic VPN use case (even tho libknet could for some areas). The worst we have seen is a roughly 30% regression on
a loopback
networking test that did a ton of syscalls and context switches.
which allegedly comes quite close as a user/kernel space mix.
Other impact exploring examples include PostreSQL[2], for example.
[1] https://lwn.net/Articles/738997/
[2]
https://www.postgresql.org/message-id/20180102222354.qikjmf7dvnjgbkxe%40ala…
--
Jan (Poki)