[128151] in North American Network Operators' Group
RE: Appliance Vs Software based routers
daemon@ATHENA.MIT.EDU (Nathan Eisenberg)
Sun Jul 25 13:01:05 2010
From: Nathan Eisenberg <nathan@atlasnetworks.us>
To: "nanog@nanog.org" <nanog@nanog.org>
Date: Sun, 25 Jul 2010 17:00:26 +0000
In-Reply-To: <4C4C653E.4030506@earthlink.net>
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
> They are all software based routers... It really shouldn't matter
> whether an Appliance Application (i.e. some routing program is running
> on a minimal runtime environment ) or a routing program is running as
> part of an OS or as an Application on an OS. It is all Software until
> it
> becomes silicon.
>=20
> The only issue is how far off the metal you are and its not hardware
> based routing really until there is no OS, no development environment,
> no software involved right?
As has been pointed out, hardware/appliance/software can be a highly semant=
ic issue, at least for some people. OP seemed like a specific question cou=
ched in vague terms - I'd rather have a discussion about what OP was trying=
to accomplish than rehash "Vyatta as a BRAS".
What's specifically important is the distinction between an 'appliance' pla=
tform (like a MIPS or Cisco routing switch), and what I presume OP infers a=
'software' platform to be (an x86 box running iptables or Quagga). In tha=
t case, I would tell OP that the PCI/PCI-e bus architecture isn't built to =
handle the rampant interrupts (or polling) that a real routing/switching wo=
rkload generates. The bus controller is built/sized to pump data to and fr=
om a video card/IO controller/etc, not to ship Ethernet packets up to the C=
PU and back out again in 8 different directions. On the other hand, moving=
packets between 8 interfaces is exactly what a routing switch like a Cisco=
3750 is built to do.
So, I wanted to retrieve the values of 'software router' and 'appliance' fr=
om OP to see if that's where he was going.
Best Regards,
Nathan Eisenberg
