Friday, May 22, 2020

Determining required buffer sizes + a sanity check

A customer of mine wants to install border devices in Equinix colocations to host 10gb wavelengths back to their campus and 10gb cross-connects to their cloud providers, who they will peer BGP with. I don't know what the exact traffic flows will look like, but I do know that there will be DC-to-CSP flows and CSP1-to-CSP2 flows.

They're a cisco shop and typically I would recommend going with ASR 1k routers for this purpose, however, the customer plans to use Nexus 3k switches instead. I don't know which ones they plan on using (they aren't procuring the equipment through us) but 3524-XL's seem like they'd make the most sense in this case. I have some reservations about this, so I'd like to get your guy's opinion on the conclusions I'm making.

  1. The nexus switches don't [appear to?] have any traffic shaping capabilities, so instead of building lower-rate virtual circuits over a shared 10gb cloud exchange connection, they're going to have to order physical cross-connects to each CSP. If not, they are going to have to oversubscribe their cloud exchange interface massively (4x 10gb VCs over a single 10GigE interface). This, in turn, may cause Equinix to indiscriminately drop packets during periods of congestion.
  2. The nexus switches have less than 10% of the buffering capabilities of an ASR 1001-X. So when traffic comes in from their 10gb wavelength circuit and goes out on a GigE cross-connect, packets could be dropped due to insufficient buffering. Therefore, all their physical cross-connects would need to be 10gb.

What do you guys think? Is my thought process sound? Am I being paranoid?



No comments:

Post a Comment