Hi guys - so most spine-leaf architectural depictions that I've seen in vendor documentation, marketing materials, and educational resources are two-tier models, with access switches connecting to every core switch in typical Clos fabric. However, it seems to me that if you're using top-of-rack switches in the access layer, the L1 infrastructure required would get a bit out of control...
If we're looking at a medium-to-large sized DC, wouldn't it make more sense to design a 3-tier model where every TOR switch connects to 2+ EOR switches, and then those EOR switches connect to all the core switches in the Clos Fabric? You could still scale-out at the access layer installing additional EOR switches if needed.
If what I'm saying DOES make sense, then why isn't this sort of design commonly depicted? I mean, I know not every deployment will use managed TOR switches; some will have EOR only, others will have FEX (though isn't Cisco sunsetting it?), but my impression has always been that TOR tends to be the preferred design choice these days, at least in pods where do you don't have a bunch of large blade servers.
No comments:
Post a Comment