Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 9 Mar 2005 00:34:13 +0100 (CET)
From:      Goran Gajic <ggajic@mail.sbb.co.yu>
To:        =?UTF-8?B?xYF1a2FzeiBCcm9taXJza2k=?= <lbromirski@mr0vka.eu.org>
Cc:        freebsd-net@www.freebsd.org
Subject:   Re: ipfilter 4.1.6 won't build on FreeBSD5.3 amd64 (fwd)
Message-ID:  <Pine.BSF.4.62.0503090016030.92805@mail.sbb.co.yu>
In-Reply-To: <422E240B.7010502@mr0vka.eu.org>
References:  <Pine.BSF.4.62.0503082118370.17320@mail.sbb.co.yu> <422E240B.7010502@mr0vka.eu.org>

next in thread | previous in thread | raw e-mail | index | archive | help
  This message is in MIME format.  The first part should be readable text,
  while the remaining parts are likely unreadable without MIME-aware tools.

--0-1056511592-1110324853=:92805
Content-Type: TEXT/PLAIN; charset=UTF-8; format=flowed
Content-Transfer-Encoding: QUOTED-PRINTABLE



On my NPE-G1 running just IOS 12.3(12a) cpu utilization was something
like 70-90% but with IOS 12.3(11)T3 it is 20% since this one has NAT=20
inside CEF and yes using small portions of address for NAT pool will=20
reduce CPU utilization and will improve NAT on 7206. However if you=20
compare prices of PC hardware and Cisco  hardware decent PC hardware with=
=20
FBSD seems like more acceptable solution to  me.  I was able to=20
bring down NPE-G1 with running simple ping -l 1000000 throu it and it
has died at ~ 80k pps, while FBSD5.3 box was able to route this=20
without any problems.

Regards,
gg.


On Tue, 8 Mar 2005, [UTF-8] =C5~Aukasz Bromirski wrote:

> Goran Gajic wrote:
>
>> Actually I was interested if Dual Opteron with FBSD5.3
>> can compare with Cisco7206 with NPE-G1 running only for NAT
>
> You'll need good motherboard, NICs, 1-2GB of RAM and quite capable
> CPU. Two won't help much, but sometimes the motherboards for two
> CPUs provide higher standard (separate buses for PCI, PCI-X slots
> instead of regular PCI etc.), so it may be beneficial, but YMMV.
>
>> purpose of some 7000 hosts (and sadly more then ~80k pps can easly bring=
 it=20
>> down and no one can comfirm that 7206 with NPE-G1 can actually process 1=
M=20
>> pps:).
>
> Yes, the 7206VXR with NPE-G1 can quite easily do 1Mpps, but the
> figures usually published are for routing. FreeBSD will also do
> this on properly configured hardware - google should return some
> useful usenet posts and discussions.
>
> 7200 is positioned as a router for ISPs, and they don't often do
> NAT - and as such, routing figures quite reliably put it in the
> 400-500kpps area (1Mpps full duplex).
>
> If Your problem lies in large NAT, either segregate the NAT process
> in few smaller chunks closer to end-users, by making few groups of
> "NAT-routers" that aggregate already NATed sessions on one main
> router, that's just routing (7200 will do just fine in that
> scenario), or buy some solution, that will do NAT in hardware.
>
> As for the 7200, if You wish, drop me an e-mail with some more
> details (running-config, exact version of IOS, modules loaded) and
> I can try to look for possible causes of poor performance. However
> please bear in mind, that NAT always requires first packet to be
> process/fast switched and some other requirements usually need to
> be met. For starters, check if You have CEF configured (`ip cef'),
> dropping all the usual Win$shit traffic (to not produce NAT
> translations for trashy traffic on the internal, ingress interface
> (via ACLs) and preferably control-plane configured - because sometimes
> DoS/semi-DoS scenarios arise from the fact, that router itself is
> slammered with packets.
>
--0-1056511592-1110324853=:92805--



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.62.0503090016030.92805>