[109964] in North American Network Operators' Group
Re: Gigabit Linux Routers
daemon@ATHENA.MIT.EDU (Marshall Eubanks)
Thu Dec 18 06:51:11 2008
From: Marshall Eubanks <tme@multicasttech.com>
To: Jeroen Wunnink <jeroen@easyhosting.nl>
In-Reply-To: <494A141F.1020707@easyhosting.nl>
Date: Thu, 18 Dec 2008 06:51:07 -0500
Cc: nanog list <nanog@nanog.org>
Errors-To: nanog-bounces@nanog.org
On Dec 18, 2008, at 4:13 AM, Jeroen Wunnink wrote:
> This might be of some use, it's a document written by one of the AMS-=20=
> IX engineers, it's a little aged (almost 2 years old) so there =20
> should be some improvement in the numbers, but it might give you =20
> some insight in the bottlenecks when pushing a Linux server to it's =20=
> max (10Gigabit in this case)
>
> http://noc.easycolocate.nl/10-GE_Routing_on_Linux.pdf
Note that this test did not involve full BGP. Given the problems that =20=
used to occur on some name
brand routers when BGP took up too much CPU, I would be careful =20
extrapolating these results if you
are planning on running full BGP. As the paper itself says,
" In a real-world situation the device might be running BGP, with a full
routing table. This will surely affect the performance of the device."
Regards
Marshall
>
>
>
>
> David Coulson wrote:
>> The boxes (3650s) came with Broadcom BCM5708 on-board, but I push =20
>> most of my traffic over these:
>>
>> 1c:00.1 Ethernet controller: Intel Corporation 82571EB Gigabit =20
>> Ethernet Controller (rev 06)
>> Subsystem: Intel Corporation PRO/1000 PT Dual Port Server =20
>> Adapter
>> Flags: bus master, fast devsel, latency 0, IRQ 58
>> Memory at c7ea0000 (32-bit, non-prefetchable) [size=3D128K]
>> Memory at c7e80000 (32-bit, non-prefetchable) [size=3D128K]
>> I/O ports at 6020 [size=3D32]
>> Capabilities: [c8] Power Management version 2
>> Capabilities: [d0] Message Signalled Interrupts: 64bit+ =20
>> Queue=3D0/0 Enable+
>> Capabilities: [e0] Express Endpoint IRQ 0
>> Capabilities: [100] Advanced Error Reporting
>>
>> There are four Intel ports in the boxes, so traffic may or may not =20=
>> stay on the same PCI-X card depending how things are flowing.
>>
>> Chris wrote:
>>> David: May I ask which NICs you use in the IBM boxes ? I see the =20
>>> Intels
>>> recommended by Mike have dual ports on one board (the docs say =20
>>> "Two complete
>>> Gigabit Ethernet connections in a single device =95 Lower latency =20=
>>> due to one
>>> electrical load on the bus").
>>>
>>
>>
>
> --=20
>
> Met vriendelijke groet,
>
> Jeroen Wunnink,
> EasyHosting B.V. Systeembeheerder
> systeembeheer@easyhosting.nl
>
> telefoon:+31 (035) 6285455 Postbus 48
> fax: +31 (035) 6838242 3755 ZG Eemnes
>
> http://www.easyhosting.nl
> http://www.easycolocate.nl
>
>
>
