Blue[1] is an Active Queue Management algorithm. Like Random early detection, it operates by randomly dropping or ECN marking packets in a router's queue before it overflows. Unlike RED, however, it requires little or no tuning on the part of the network administrator.
Operation of Blue
A Blue queue maintains a drop/mark probability p, and drops/marks packets with probability p as they enter the queue. Whenever the queue overflows, p is increased by a small constant d, whenever the queue is empty, p is decreased by a constant i<p.
Stochastic Fair Blue
The main flaw of Blue, which it shares with most single-queue queing disciplines, is that it doesn't distinguish between flows, and treats all flows as a single aggregate. Therefore, a single aggressive flow can push out of the queue packets belonging to other, better behaved flows.
Stochastic Fair Blue (SFB)[2] is a stochastically fair variant of blue which hashes flows and maintains a different mark/drop probability for each hash value. Assuming no hash collisions, SFB is able to provide a fair share of buffer space for every flow. In the presence of hash collisions, SFB is only stochastically fair.
Unlike other stochastically fair queuing disciplines, such as SFQ, SFB can be implemented using a Bloom filter rather than a hash table, which dramatically reduces its storage requirements when the number of flows is large.
When a flow's drop/mark probability reaches 1, the flow has been shown to not react to congestion indications from the network. Such an inelastic flow is put in a penalty box, and rate-limited.