Bulk synchronous parallel computing

Since 1944 von Neumann's model for sequential computing has been widely accepted, but there is no standard model for parallel computing. Most approaches nowadays are based on message-passing, but they are often inadequate, since the potential danger of deadlock, in which each possible activity is blocked, waiting on some other activity that is also blocked, increases dramatically with the complexity of software.

Furthermore, models based on message-passing, e.g. MPI (Message passing interface), do not easily allow performance prediction. The Bulk synchronous parallel computing model (BSP) however abstracts low-level program structure in favour of so-called supersteps. This allows easy debugging, removes the problem of deadlock and allows a reasoning of the correctness of the code nearly as easily as in sequential code.