On Wed, 2021-06-23 at 20:07 +0200, Clemens Zauner wrote: > > And what about the output of : > > sysctl net.route.algo > > [root@amarklor]/etc# sysctl net.route.algo > net.route.algo.debug_level: 5 > net.route.algo.inet.algo: dpdk_lpm4 > net.route.algo.inet.algo_list: dpdk_lpm4, bsearch4, radix4_lockless, > radix4
Ah I see; thats a new forwarding scheme. Is it possible to switch back
to radix4(_lockless) in a more or less hitless fashion (this is a
Boarder-router, full table, forwarding 'real' traffic)?
If yes, I give It a try - not during main-time, but in the evenings.
If I require a reboot ore something disruptive , I'm willing to give It
a try; but I have to engineer an excuse for a short-notice-troube-
ticket ('Increased sunspot activity', or something along that way).
And, as I'm on my 'asking stupid questions' - train: what is the
difference between radix4 and radix4_lockless? I got the memo, that
bsearch(4) is *not* the way to go for large routing-tables, but I'm
somwhat lost regarding the difference between the 'normal' and
'_lockless' one.
Yes, I saw that radix_* is slower, especially also due to high cahce-
usage. But: better slow and functional. Besides: Epyc System here;
slightly larger L3-Cache than the Core-i7 which was used for the
benchmarks I found online.
Multipath-routing is not a requirement for me. I can handle my flows
without residing to multipathing (assuming that means multiple active
gateways for the same destinations).
As this seems to be turned on here:
[root@amarklor]/etc# sysctl net.route.multipath
net.route.multipath: 1
cu
Clemens.
smime.p7s
Description: S/MIME cryptographic signature
_______________________________________________ Bsdrp-users mailing list Bsdrp-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bsdrp-users
