The MPLS-OPS Archive

Cell Relay Retreat>MPLS-OPS Archive>month:2004-Aug> msg00123



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

RE: MPLS aware NICs

  • From: "Bell, John" <john.bell@thus.net>
  • Date: Fri, 20 Aug 2004 14:50:38 +0100
  • Cc: "McCallum, Robert" <robert.mccallum@thus.net>, "'Puddinhead Wilson'" <puddinghead_wilson007@yahoo.co.uk>, mpls-ops@mplsrc.com
  • Resent-Date: Fri, 20 Aug 2004 10:10:29 -0400
  • To: Bjørn Mork <bjorn@mork.no>, "Bell, John" <john.bell@thus.net>
  • X-MIME-Autoconverted: from quoted-printable to 8bit by host.secure4-hosting.net id i7KDpUgJ006950

Why not indeed? Except I'd terminate the MPLS tunnel on a router that serves
the server farm, and pass the IP packets alone to the server. Why push the
load onto a server that has its own jobs to do? You'll go far in management
with that attitude, my lad! ;-)And besides most server high-availability
software is designed for IP alone. I'm assuming, of course, that this server
farm has a secure connection to its own LAN router. And the label is only of
use when there are switching points ahead. Once you're onto the last leg of
the trip, there's no need for the label. I still think it makes most sense
to keep MPLS in the routing domain. And leave the poor, maxed-out server
serving!

Cheers, 
	John

PS But if you really want to do it anyway, why wouldnt a standard Ethernet
NIC handle it? So long as your frame aint too big, it shouldnt care!

-----Original Message-----
From: Bjørn Mork [mailto:bjorn@mork.no]
Sent: Friday, August 20, 2004 12:33
To: Bell, John
Cc: McCallum, Robert; 'Puddinhead Wilson'; mpls-ops@mplsrc.com
Subject: Re: [MPLS-OPS]: MPLS aware NICs


"Bell, John" <john.bell@thus.net> writes:

> 	None that I'm aware of. As Steinar says, you'd have to run a routing
> protocol on your workstation/server for it to make any sense. Equally,
> making an NIC MPLS-aware means writing an MPLS stack in software, so I
> suppose  some guys in a Uni lab may write their own MPLS-aware NIC stack
as
> a project, and run "routed" on a Linux box, just to understand whats
> happening. But there will never be a commercial reason to do it, as you
need
> to turn your server into a software router. As we all know, this is a
crappy
> idea when there are perfectly good (i.e a million times better) hardware
> routers around :-)

Let's say you want to do policy routing on your edge, redirecting some
packets to one or more servers located at a central server farm.  To
do this you need tunnels from the edge routers to these servers.  Why
not use MPLS tunnels?  The servers will have to terminate the tunnels
no matter what protocol is used.  But they don't have to do any
forwarding, and they don't really have to participate in the routing
exchange.

I'd say that popping a label is much preferred to unwrapping ipsec or
gre or whatever, even if you don't qualify as a router ;-)


Bjørn

-------
The MPLS-OPS Mailing List
Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
Archive: http://www.mplsrc.com/mpls-ops_archive.shtml