SD-Access Transit Node Scalability

[newtonr]

Regarding the sizing of an SD-Access Transit Control Plane node, will a pair of Cisco 9300 switches be sufficient for a 3-site design (hospitals) where there would be anywhere between 4000 and 6000 active endpoints per site at any one time and approx 50 fabric edge switch stacks and 400 fabric APs per site? 

Based on the datasheet, the C9300 is suitable for SDA up to 16,000 endpoints and 8,000 IPV4 routes but I assume this is when used as a Site-Local Control-Plane Node that requires EID-to-RLOC mappings. With the Transit CP Node, the CVD suggests the mappings are 'site aggregate prefix registrations' that create an aggregate HTDB for all fabric sites connected to the transit.

Does this mean that only the /32 RLOC aggregates are forwarded to the Transit CP, which works out to be all fabric edge nodes in all sites, equating to 150 RLOCs, or would it be all endpoint mappings from all sites? Can't seem to find the correct answer I'm looking for but hoping it's just FE RLOCs as this means the 9300 would easily be suitable.

Cheers,

Rob

[jedolphi · Accepted solution]

Hi Rob. The SD-Access Transit CP holds summary routes for access networks at fabric sites, not host routes e.g. if you provision 10.0.0.0/21 into CORP VN at fabric site 1, then regardless of whether there is 1 or 5000 endpoints in the 10.0.0.0/21 network, the Transit CP will have a single entry, the summary route of 10.0.0.0/21. On that basis, the number of endpoints, access switches and fabric-enabled wireless APs has no impact on the load placed on the SD-Access transit CP. If economics is a primary concern then you may want to check the price of some other platforms for same role e.g. ISR4K. Cheers, Jerome

[newtonr]

Thanks for the confirmation Jerome. It backs up what I thought was the case. Cheers.

Rob