Friday, June 14, 2019

Spine/Leaf - Overlay necessary?

Greetings,

Been a long time lurker here, and recently took (was voluntold to take) a more active role in network administration at my job. There may have been an RSTP issue a few weeks back that caused a production outage, and I am the only one who has some room in my schedule to address this. So I am somewhat new and trying to take a slow and structured approach to this.

I'm tasked with figuring out how to move up from an older Juniper design we have been limping along. We have 8 racks with virtual chassis EX ToRs, each has redundant aggregate uplinks to a 2 member virtual chassis 10g EX distribution switch, and finally a firewall cluster north of that. We only have 4 production VLANs plus I think 2 test VLANs, those gateways are on the firewalls.

Most diagrams I've seen seem to have two spines minimum, so I'm wondering if it would be worth it to split the distribution VC into separate spine switches. Assuming I do that, and go with eBGP on each link and private ASN per switch to put the L3 gateways on the leafs, my question is whether it's necessary for an overlay like VXLAN or if just eBGP is enough since we aren't a large or complex DC.

We have some home grown IPAM system and needless to say, huge chunks of the 10.x.x.x and 192.168.x.x networks available for IP ranges. Initially I was thinking of carving out eight /24's to represent the first prod VLAN, one /24 per leaf with the gateway assigned to an irb interface, and repeat for each additional VLAN. That also seems like a LOT of IPs to reserve so then I started considering using /25's to slim things down but I was concerned if that would over complicate this situation.

Or just scrap BGP altogether and try OSPF. I like the idea of AS prepending though to bleed off traffic for maintenance though.



No comments:

Post a Comment