As we all know, Giraph is based on the BSP parallel computation model. My
question to Giraph developers is that what are the features or key
advantages of Giraph that were *only possible* because of the BSP model?

One aspect is scalability. Does the BSP model fundamentally contribute to
Giraph's scalability? How? One example is enabling the use of combiners
(because messages are sent at the same phase and hence can be combined),
but what else did you implement and relied on the assumption of BSP to make
Giraph more scalable?

Another aspect is usability. One benefit of the BSP model is touched by
Giraph users, as they only need to think in terms of phases of local
computation and network communication, which is simple enough. Do Giraph
developers get any advantage like that?

Reply via email to