2:29 p.m.
Hi Magnus,
I have discussed the DCN design with colleagues from NetEng and R&D team.
Please have a look at attached draft of DCN overview: dcn.overview.3loops.jpg. This is the
setup we would like to recommend for your network:
- optical rings below 32 NEs
- no RSTP enabled on electrical ports
- we have less subnets
- we could split red LAD into two subnets to limit the problem to smaller segment in case
of fiber cut
We have done this number of times in such way. I know from practice that this is the most
stable DCN we can get.
If we would go for version you have suggested: DCN-no-l2-loops.pdf you would be the first
one to test it. In theory it should work as well. However we would know it only after
adding all the nodes and having network at full capability.
Please be aware that this is not something we would usually suggest to customer.
In case of fiber cut in such network convergence times will be counted in minutes (around
2 min due to OSPF counters settings) in compare to seconds when RSTP is used.
We need to take into account that worst case scenario is that R&D will suggest DCN
redesign.
When it comes to OSPF area you will have at least 115 POP sites + 13 active UNI sites (128
total). We also need to take into account possibility that the network might extend in the
future. R&D wasn't that optimistic to have such amount of NEs in single area.
Tests we have performed during the POC only show that NCU can process 10.000 LSA entries.
We need to take into account that packets must be passed correctly (e.g. some ping test
could have been executed) and that network might work in some overload/unexpected
conditions (e.g. link flapping caused by fibercut, faulty hardware in the network). On top
of that LSA exchanged to share traffic engineering information are much bigger to usual
once for subnets and hosts information exchange. Due to that having single database
storing all that data might cause problems.
Taking all above into account I think we should divide your network into two OSPF areas.
Please let me know what you think about our suggestions.
--
BR
Michal
-----Original Message-----
From: Magnus Bergroth [mailto:bergroth at nordu.net]
Sent: Thursday, November 05, 2015 11:32 PM
To: Michał Szpilewicz <MSzpilewicz at advaoptical.com>; "Björn Lindman
(bjorn.lindman at ipnett.com)" <bjorn.lindman at ipnett.com>
Cc: Magnus Bergroth (bergroth at sunet.se) <bergroth at sunet.se>; Ari Salomaa
<ASalomaa at advaoptical.com>; Torbjörn Rium <trium at advaoptical.com>;
Christiane Hack <CHack at advaoptical.com>; Stefan Josefsson <stefan.josefsson at
ipnett.com>; sunetc-teknik at lists.sunet.se
Subject: Sunet Adva DCN
Hi Michal and Björne
As we talked to day regarding the DCN design I have now drawn an alternative DCN to show
how we would like to have the DCN.
Criteria,
Only one ospf area
No Layer2 loops
Simple
It ended up with three more broadcasts domains.
The suggestion needs two more OSCM cards than the original design one i Malmö and one in
Rätan both has 9RU shelfs so I guess they would fit.
With some more thoughts it would probably be possible to design it so that it could be
divided into two ospf areas, without l2 loops. But we prefer to have every thing i area
0.
Would it be possible to build like this and would it have any draw backs?
<<Magnus