To support this growth, data center network architectures have evolved into scale-out designs, providing the plumbing to carry large amounts of latency-sensitive application traffic between racks. To improve bandwidth utilization, data centers are also moving from 10GbE to 40GbE to 100GbE, providing bigger pipes for transmitting application traffic at line rates. At the same time, more and more servers are being packed into data centers
Multi-path forwarding has been deployed extensively in data center networks as a way to increase capacity and resiliency. Flow-based hashing is widely used to split traffic onto multiple paths. However, such method may work poorly when there is a small number of disproportionaly large flows exists. This post will explore a dynamic load balancing mechanism called daptive flowlet splicing (AFS).
The terminologies of “single-tier” or “two-tiered” fabric have been used to describe network fabric or cluster-based products. I’m doubtful whether a two-tiered fabric can truly function as a single-tier network, but certainly not all two-tiered networks are made equal.