[afnog] afnog Digest, Vol 54, Issue 23

[Peter Nyamukusa]

Hi Mark,

Points noted & taken for me I try to keep things basic and as simple as
possible so that it will be easy to explain & show a new network engineer.
I don't want to be adjusting timers and other configurations which will make
his life a nightmare especially in Africa where you know good hands are hard
to come by (less Admin work for me).
I agree with you in designing a hierarchical network and we also need to
look at designing a fully converged network

Just my 2 cents

Cheers,
Peter.


-----Original Message-----
From: Mark Tinka [mailto:mtinka at globaltransit.net] 
Sent: Wednesday, September 24, 2008 2:47 PM
To: Peter Nyamukusa
Cc: 'Frank A. Kuse'; afnog at afnog.org
Subject: Re: [afnog] afnog Digest, Vol 54, Issue 23

On Wednesday 24 September 2008 18:07:04 Peter Nyamukusa
wrote:

>     *  IS-IS does not use level 3 routers.

Well, L3 routing is inter-domain. OSI's approach was IDRP.

> * Unlike OSPF, IS-IS routers are not required to be connected to a 
> contiguous backbone area. In fact, the backbone area can also be 
> segmented in IS-IS.

A major advantage when expanding your network beyond multiple regions.

>     * With IS-IS, there is no restriction that all backbone routers 
> (level 2 routers) be contiguous
>       such as the backbone area of OSPF.

Actually, there is... 

If you deploy a hierarchical IS-IS network (L1 routers within the PoP's, L2
[or rather, L1/L2] routers in the core), your L2 backbone MUST be
contiguous.

The merit is that L1 routers don't need to be contiguous between PoP's.
Route leaking ensures optimal routing (longest match) between PoP's.

>     * In OSPF all areas must be directly linked to area 0, and the 
> backbone area must also not be segmented.

A problem for networks that build "wide".

>     * With IS-IS, the backbone area can be more easily extended since 
> all L2 routers need not be linked directly together.

Actually, it's the other way around; L1 routers do not need to be linked
directly.

However, L2 or L1/L2 routers should be contiguously deployed. L2 or L1/L2
"snake" their course in between L1 routers across the various regions an ISP
may deploy its network.

>     * With regard to CPU use and the processing of routing updates, 
> IS-IS is more efficient
>       than OSPF.

IS-IS is fundamentally less chatty than OSPF on the wire. 
Both IGP's have several knobs and switches that can be manipulated to make
the deployment even more scalable, e.g., the LSP lifetime can be configured
up to 18.2hrs in IS-IS, which makes it quite suitable for large deployments,
e.t.c.

It's also good to note that IS-IS was easily extendable for the inclusion of
new features, e.g., IPv6 routing, as it utilizes TLV's, rather than OSPFv2
which uses the "Options" 
field (already depleted) - hence the transition to a new OSPF protocol,
OSPFv3.

Suffice it to say that 'draft-ietf-ospf-af-alt-07.txt' 
(expired the last time I checked) attempts to extend OSPFv3 by adding
multiple address families, one of which is IPv4 (Unicast and Multicast). Of
course, it runs over IPv6, so an IPv6 network would be a requirement for
running IPv4 over OSPFv3. Vendor support has already emerged from Juniper...
but I digress.

And then of course, IS-IS runs directly over the data link layer.

> Try to design your network so it can run in a single area/level which 
> makes things like TE/FRR...

I believe a hierarchical network design provides better scalability.

Besides, I'd design my IS-IS backbone with IP in mind. Other features like
MPLS-TE and FRR are secondary (as they depend on a stable IP foundation).

However, it is true that inter-area CSPF for MPLS-TE does not exist today
(major vendors have planned to support it, though). LSR's have to be
'loosely' defined in the path list for RSVP to successfully signal an
inter-area LSP.

We, generally, treat MPLS-TE as a tactical response to a potential
congestion problem, rather than a lasting solution, especially since very
few points would be affected in the grand scheme of things... so manual
inter-area MPLS-TE isn't an issue.

Cheers,

Mark.