Axiomatization and characterization of BSP algorithms

Axiomatization and Imperative Characterization of Multi-BSP Algorithms: A Q&A on a Partial Solution

“…In [20], we extended these results to bsp and proved that algo bsp =asm bsp ≃ imp bsp . To do so, we add another postulate that arranges the steps into bsp's supersteps and in a way that the bsp algorithms no longer work on a single memory (a first-order structure) but on p-tuples of arbitrary sizes, on per computing unit.…”

“…In particular, these algorithms must work for any number of computing units (up to some constraints such as a number of processors equal to a power of two). This raises two questions: (1) can we characterize the class of the multi-bsp algorithms so that it is a natural extension of the class of the bsp algorithms [20]? And (2), can we prove that all the multi-bsp algorithms can be programmed using a programming language?…”

“…Furthermore, the study of class models [14,19,20] allows us to classify what can or cannot be effectively programmed. Indeed, if it is known that mainstream languages such as c or java are turingcomplete, which means that they can simulate the computation of any computable function (so the inputoutput of any turing machine, up to an unbounded memory), what is called algorithmic completeness rather focuses on the step-by-step behavior that is, a model could compute all the desired functions, but some algorithms (ways to compute these functions) could be missing [14].…”

“…how to perform integer additions or how to manage the communications). Because the multi-bsp model is ("just") a hierarchical extension of bsp, it seems appropriate to ask the question of how to extend the previous results about bsp [20] in a "minimal" way rather than doing all the proofs from scratch. We will show that this is mostly feasible but that there are some points that require more important modifications due to the use of nested supersteps.…”

Electric vehicle regional management system based on the BSP model and multi-information fusion

“…For the Shared storage mode, the remote read operation will be deemed as two messages passing, and the influence of pipeline prefetching technology, cache-induced data inconsistency and Cache hit ratio on the calculation is not considered, which will bring data deviation to the system designed in this paper. In the present work, the BSP model (Allombert & Gava, 2020;Liu et al, 2020;Marquer & Gava, 2019) adopts obstacle synchronization mode to realize global synchronization by hardware, thus providing an effective way to execute tightly coupled synchronous parallel algorithm, and it's adopted to divide m supersteps into m EV, which can effectively avoid the deadlock phenomenon.The fence can be synchronized according to the different lengths of the three signals in each superstep.…”