netdev - Re: [RFC bpf-next 0/7] busy poll support for AF

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Fri, 17 May 2019 11:20:03 -0700
From:   Jakub Kicinski <jakub.kicinski@...ronome.com>
To:     Magnus Karlsson <magnus.karlsson@...il.com>
Cc:     Alexei Starovoitov <alexei.starovoitov@...il.com>,
        Magnus Karlsson <magnus.karlsson@...el.com>,
        Björn Töpel <bjorn.topel@...el.com>,
        Daniel Borkmann <daniel@...earbox.net>,
        Network Development <netdev@...r.kernel.org>,
        bpf@...r.kernel.org, Jonathan Lemon <bsd@...com>,
        Maciej Fijalkowski <maciejromanfijalkowski@...il.com>,
        "Samudrala, Sridhar" <sridhar.samudrala@...el.com>
Subject: Re: [RFC bpf-next 0/7] busy poll support for AF_XDP sockets

On Thu, 16 May 2019 14:37:51 +0200, Magnus Karlsson wrote:
>                                       Applications
> method  cores  irqs        txpush        rxdrop      l2fwd
> ---------------------------------------------------------------
> r-t-c     2     y           35.9          11.2        8.6
> poll      2     y           34.2           9.4        8.3
> r-t-c     1     y           18.1           N/A        6.2
> poll      1     y           14.6           8.4        5.9
> busypoll  2     y           31.9          10.5        7.9
> busypoll  1     y           21.5           8.7        6.2
> busypoll  1     n           22.0          10.3        7.3

Thanks for the numbers!  One question that keeps coming to my mind 
is how do the cases compare on zero drop performance?

When I was experimenting with AF_XDP it seemed to be slightly more
prone to dropping packets than expected.  I wonder if you're seeing
a similar thing (well drops or back pressure to the traffic generator)?
Perhaps the single core busy poll would make a difference there?