netdev - Re: [PATCH v15 ] net/veth/XDP: Line-rate packet forwarding in kernel

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:   Wed, 4 Apr 2018 17:09:01 -0400
From:   "Md. Islam" <mislam4@...t.edu>
To:     Jesper Dangaard Brouer <brouer@...hat.com>
Cc:     netdev@...r.kernel.org, David Miller <davem@...emloft.net>,
        David Ahern <dsahern@...il.com>,
        Stephen Hemminger <stephen@...workplumber.org>,
        Anton Gary Ceph <agaceph@...il.com>,
        Pavel Emelyanov <xemul@...nvz.org>,
        Eric Dumazet <edumazet@...gle.com>,
        alexei.starovoitov@...il.com
Subject: Re: [PATCH v15 ] net/veth/XDP: Line-rate packet forwarding in kernel

On Wed, Apr 4, 2018 at 2:16 AM, Jesper Dangaard Brouer
<brouer@...hat.com> wrote:
>
> On Sun, 1 Apr 2018 20:47:28 -0400 Md. Islam" <mislam4@...t.edu> wrote:
>
>> [...] More specifically, header parsing and fib
>> lookup only takes around 82 ns. This shows that this could be used to
>> implement linerate packet forwarding in kernel.
>
> I cannot resist correcting you...
>
> You didn't specify the link speed, but assuming 10Gbit/s, then the
> linerate is 14.88Mpps, which is 67.2 ns between arriving packets. Thus,
> if the lookup cost is 82 ns, thus you cannot claim linerate performance
> with these numbers.
>
>
> Details:
>
> This is calculated based on the the minimum Ethernet frame size
> 84-bytes, see https://en.wikipedia.org/wiki/Ethernet_frame for why this
> is the minimum size.
>
> 10*10^9/(84*8) = 14,880,952 pps
> 1/last*10^9    = 67.2 ns
>

Yes, it's not actually line-rate forwarding, but it shows the intent
towards that. Currently we are doing many things in fib_table_lookup()
that can be simplified for a router. fib_get_table() and FIB_RES_DEV()
would be simplified if we disable IP_ROUTE_MULTIPATH and
IP_MULTIPLE_TABLES. We can increase throughput by doing less :-)
Moreover if a network mostly carries larger packets (for instance, a
network exclusively used for video streaming), then a 40Gb NIC
produces packets in every 300ns.

40*10^9/(1500*8) = 3.4mpps
1/last*10^9    = 300 ns

> --
> Best regards,
>   Jesper Dangaard Brouer
>   MSc.CS, Principal Kernel Engineer at Red Hat
>   LinkedIn: http://www.linkedin.com/in/brouer