The evaluation section (in the blog) talks about "traditional data centre networks." What are they?
Cisco has released a data centre network design that has been out for nearly 6 years now. The design outlined in the VL2 paper is similar.
@those who operate DC networks: what does your topology look like? How do you handle oversubscription in a fair manner? If anyone here works at Amazon, could they shed light on fairness issues: can one VM blasting traffic using UDP hurt another that uses vanilla TCP?
It looks like this guy just reimplemented the worst parts of ATM (directory lookups don't scale) without any of the good parts (predictable bandwidth guarantees).
Cisco has released a data centre network design that has been out for nearly 6 years now. The design outlined in the VL2 paper is similar.
@those who operate DC networks: what does your topology look like? How do you handle oversubscription in a fair manner? If anyone here works at Amazon, could they shed light on fairness issues: can one VM blasting traffic using UDP hurt another that uses vanilla TCP?