[j-nsp] Inter-racks switch routing recommended practice

[Wayne Tucker]

You should be able to run that many racks inside a single OSPF area -
in fact, multiple areas can result in a lot of type 3 LSAs if you do
not summarize properly.  You can improve initialization times and keep
the LSDB down to one LSA per router if you:

1.) Set the RVIs on the ToRs to passive (to keep type 2 LSAs from
being generated for the host subnets)
2.) Configure all of the ToR<->EX4550 links as point-to-point (ditto,
plus avoid the DR election delays)

At larger scales the frequency of LSA refreshes and SPF runs would
make multiple areas (or other solutions) worthwhile, but for these
platforms you're probably looking at hundreds of racks (or lots of
really flaky links ;) before that even begins to be a conern.

I can't think of anything BGP would provide that would be of
significant benefit based on what you've described - plus I believe it
requires additional licensing on those platforms.

:w




On Tue, Jun 4, 2013 at 7:24 PM, Ihsan Junaidi Ibrahim <ihsan at grep.my> wrote:
> Hi,
>
> I'm building an infrastructure which comprises of a few tens of racks with Hadoop, Supermicro MicroCloud and whatnot running. Each rack probably will have EX4200 or EX3300 ToR switch, individually at the moment, not VC-chained. These switches will have a couple of EX4550 aggregating the circuits.
>
> My question is what would be the best routing protocol in this kind of scenario?
>
> I'm thinking multi-areas OSPF/v3 but would a flat OSPF area 0 topology with BGP make more sense? I don't have a lot of exposure in dense datacenter routing so I'm bringing the conventional WAN routing thinking cap into the picture.
>
> Thanks.
>
>
> _______________________________________________
> juniper-nsp mailing list juniper-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/juniper-nsp