netdev - Re: I can tell no FIB

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]

Message-ID: <54F8EA61.9010706@redhat.com>
Date:	Thu, 05 Mar 2015 15:44:33 -0800
From:	Alexander Duyck <alexander.h.duyck@...hat.com>
To:	Dave Taht <dave.taht@...il.com>, Scott Feldman <sfeldma@...il.com>
CC:	Netdev <netdev@...r.kernel.org>
Subject: Re: I can tell no FIB

On 03/05/2015 01:01 PM, Dave Taht wrote:
> On Thu, Mar 5, 2015 at 12:49 PM, Scott Feldman <sfeldma@...il.com> wrote:
>> Hi Alex, turns out you're required to take a mandatory week-long
>> vacation after your fourth patch set to net/ipv4/fib_*.  See you in a
>> week!  Take lots of pictures.
> +10!
>
> Is anyone here working on testing the new FIB stuff on itty bitty 32
> bit platforms? It looks really promising
> but openwrt is stabilizing on 3.18 and the prospect of backporting all
> this stuff to that to test at scale is intimidating. (*I* am willing
> to wait for 4.2)
>
> But boy, could openwrt test at scale:
>
> https://downloads.openwrt.org/snapshots/trunk/

As far as what to backport I would recommend only targeting the FIB 
changes that went into 4.0.  It was low risk for causing regressions and 
significant benefit.  This is why I referred to them as "low hanging 
fruit" when I described them  in my presentation at Netdev 0.1.  The 
stuff that went into 4.0 reduced things by hundreds of nanosecnds in 
some cases, the stuff targeting net-next/4.1 is only going to reduce 
things by tens of nanoseconds.

What I am working on now is basically just trying squeeze the last bits 
of performance out of what is left.    Excluding the main/local merge I 
have been able to remove 20% (10 - 35ns depending on the test) of the 
remaining CPU overhead for the fib table look-up with what has been 
submitted since net-next reopened.  The local/main trie merge pushes 
that to somewhere around a 40% reduction from what I have seen.

- Alex
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html