A Parallel Virtual Machine for Bulk Synchronous Parallel ML

Bulk Synchronous Parallel ML or BSML is an extension of ML for programming direct-mode parallel Bulk Synchronous Parallel algorithms as functional programs. Bulk-Synchronous Parallel (BSP) computing is a parallel programming model introduced by Valiant [20] to offer a high degree of abstraction like PRAM models and yet allow portable and predictable performance on a wide variety of architectures. A BSP algorithm is said to be in direct mode [5] when its physical process structure is made explicit. Such algorithms offer predictable and scalable performances and BSML expresses them with a small set of primitives taken from the confluent BSλ calculus [12]: a constructor of parallel vectors, asynchronous parallel function application, synchronous global communications and a synchronous global conditional. Our BSMLlib library implements the BSML primitives using Objective Caml [10] and MPI [19]. It is efficient [11] and its performance follows curves predicted by the BSP cost model (the cost model estimates parallel execution times). This library is used as the basis for the CARAML project, which aims to use Objective Caml for Grid computing with, for example, applications to parallel databases and molecular simulation. In such a context, security is an important issue, but in order to obtain security, safety must be first achieved. An abstract machine is used for the implementation of Caml and is particular easy to prove correct w.r.t. the dynamic semantics [6]. In order to have both simple implementation and cost model that follows the BSP model, nesting of parallel vectors is not allowed. BSMLlib being a library, the programmer is responsible for this absence of nesting. This breaks the safety of our environment. A polymorphic type system and a type inference has been designed and proved correct w.r.t. a small-steps semantics.