[80525] in North American Network Operators' Group
Re: Slashdot: Providers Ignoring DNS TTL?
daemon@ATHENA.MIT.EDU (Michael.Dillon@radianz.com)
Tue May 3 09:55:25 2005
In-Reply-To: <20050501082207.GM696@overlord.e-gerbil.net>
To: nanog@merit.edu
From: Michael.Dillon@radianz.com
Date: Tue, 3 May 2005 14:57:24 +0100
Errors-To: owner-nanog@merit.edu
> 7200 most certainly does not have interface processors. 7500 does have
> processors on the VIPs that do forwarding lookups in a distributed
> fashion, but the same procedure for software forwarding apply, there
just
> happen to be a few more CPUs floating around.
Also, it may not be clear from a manufacturer's literature just
how many CPUs are in a router. FPGAs and other ASICs can include
one or more processor cores in the custom logic. A modern router
is actually a box containing a network of specialized computers
interconnected with various high-speed network fabrics to do
jobs like move packets and internal coordination of activities.
Routing engineers tend to have a higher level abstract concept
of what the router does, based on older (less modern) implementations
which implemented new concepts like CEF, etc.
> The original poster is entirely correct. The original Internet Processor
> is still supported, but it is about as far from "widely used" as you can
> get. If you're still using it you have bigger problems. The IP2 is only
> capable of doing per-flow load balancing, which is probably a good
thing.
And, if the few older models in existence are actually in a
PPLB configuration, it is highly likely that they will only
be dealing with load balancing over n identical circuits with
between point A and point B in which all the A ends are on the
same router and all the B ends are on the same router. This happens
to be the scenario in which PPLB doesn't trigger major problems.
> Nothing says that you can't have out of order packets on the Internet
> until you are blue in the face. However, it tends to do very nasty
> things to the TCP algorithm, which makes it perform poorly.
Which is precisely the reason why PPLB is not an issue with
anycasting. People who tried PPLB got burnt and got rid of
the config that was causing the pain. Therefore, per packet load
balancing is extremely rare in the Internet today, and getting rarer.
Now can we leave the discussion of theoretical problems behind?
If anything, this should be taken up at the IETF so that the
downsides of per packet load balancing are publicized and PPLB
can be deprecated.
--Michael Dillon
