@nickrusso42518@SharpNetwork I would also say that this swings both ways. If you’ve never stuck around long enough to see how your greenfield designs end up eventually breaking in unexpected ways, you’re bound to repeat the same mistakes at the next gig.
@nickrusso42518 Many engineers love solving hard problems with nerd knobs. While once in my career I was the same way, I now find it less and less interesting to work on those types of networks (knowing with full certainty how they end up in the end).
@ryanmerolle@dmfigol Yeah, that was my initial thought, but I can imagine that the length of these contexts would get out of hand once you apply more than a few rules.
For those that rely on Netbox as their source of truth, what have you done in terms of representing ACLs and prefix lists? Bend Netbox to somehow represent them or store your ACL/prefix list truth elsewhere?
@awareness_tiger 3:1 is usually a safe “set it and forget it” ratio for typical workloads. It’s when you get to bursty or more demanding ones, where you have to plan for more.
@awareness_tiger You're basically talking about oversubscription rate. In your case, if you want 1:1 non-blocking and cannot support 20 servers without 200G of uplink, you either decrease the number of servers or accept the fact that oversubscription won't be 1:1 (very common)
Has anyone come across a leaf/spine calculator before, where you feed it various inputs (# of ports, oversub ratio, etc)? Figured I would ask others, before I begrudgingly attempt to build one myself
@networkservice I remember speaking to a large enterprise last year. When discussing roadmap and "planning-ahead", we (the vendor) asked them what missing features they'd like to see. Their answer was "the same thing we tell all our vendors: the feature we're most interested in is less bugs".
@andycies@adamlevitan I did see that, but got the impression that none of the ETR guys were remotely interested in him for cash, based on the Friday night show discussion.