
Quorum is quite a well-known concept that allows for tunable consistency. But then, it is also weak at (mostly) inter-datacenter partition.
Imagine you have distributed system with replication factor of 5, and you have 3 replicas in datacenter A and 2 in datacenter B. What if parition occurs as follows?:

Turns out there is no easy solution. You just need to hope that the server is connecting to the cluster where quorum is met, namely A in this case.
N = 5R + W > N, let's give W = 3 and R = 3. When you go for strict quorum, it just fails if required quorum is not achieved.
N = 5Essentially, all write and read requests go to preferred replicas.

If one of the preferred replica goes down, however, it sends the request to secondary replica:

Cluster manager then, when failed node(R3) comes back, let the node(R4) which has received data on behalf transfer the values back to now resurrected node.

This tranferring process is called Hinted Handoff.
Now, with the same number of nodes and quorum requirements, let's compare strict and sloppy quorums.
Availability
W nodes are available.Consistenty
Performance
Complexity
hinted hadoff)automatically choosing preferred replicas? phi-accural could be applied. Spanning Tree is also a good option that could represent the most efficient, uni-directed graph.