Delivered extremely well. Thanks for sharing! How do you manage congestion & routing engineering decisions? I presume they would be either using DCTCP or similar approaches with controllers pushing flow rules from somewhere. If so how does controllers play into this topology/redundancy? (Just another/multiple servers attached to a TOR?)
Fabric is a generically used term to indicate the base level of DC Ops. Usually, this means racking, power and network topology and config. Depending on your level of automation it may also include initial OS installs to enable config. management to take over and turn the machine into what it needs to be.
it looks like VXLAN overlay, however it seems like they're doing static routes for underlay as BGP is only routing protocol, p.s. facebook have their own white boxes they are making their own rules
Alexey. Good to see your face.
Delivered extremely well. Thanks for sharing! How do you manage congestion & routing engineering decisions? I presume they would be either using DCTCP or similar approaches with controllers pushing flow rules from somewhere. If so how does controllers play into this topology/redundancy? (Just another/multiple servers attached to a TOR?)
Ok
Looks like a neural network
Cool.
what do you mean by fabric ? Is it fabric path , qfabric ? because these technologies are not interoperable , So you are stuck with a vendor for them.
Fabric is a generically used term to indicate the base level of DC Ops. Usually, this means racking, power and network topology and config. Depending on your level of automation it may also include initial OS installs to enable config. management to take over and turn the machine into what it needs to be.
it looks like VXLAN overlay, however it seems like they're doing static routes for underlay as BGP is only routing protocol, p.s. facebook have their own white boxes they are making their own rules
Clos
it's called spine and leaf, ah did you even listen to this?
@@wiziek Thanks for replying to a 7-year-old comment with your useless input