Supporting dynamic parallel object arrays

Abstract

SUMMARYWe present efficient support for generalized arrays of parallel data driven objects. Array elements are regular C++ objects, and are scattered across the parallel machine. An individual element is addressed by its 'index', which can be an arbitrary object rather than a simple integer. For example, an array index can be a series of numbers, supporting multidimensional sparse arrays; a bit vector, supporting collections of quadtree nodes; or a string. Methods can be invoked on any individual array element from any processor, and the elements can participate in reductions and broadcasts. Individual elements can be created or deleted dynamically at any time. Most importantly, the elements can migrate from processor to processor at any time. We discuss support for message delivery and collective operations in the face of such dynamic behavior. The migration capabilities of array elements have proven extremely useful, for example, in implementing flexible load balancing strategies and for exploiting workstation clusters adaptively. We present the design, an implementation, and performance results.

Exploring Traditional and Emerging Parallel Programming Models Using a Proxy Application

“…CHARM++ is a parallel programming system based on message-driven migratable objects [10], [25]. It is implemented as additions to the C++ language coupled with an adaptive runtime system.…”

Controlling Concurrency and Expressing Synchronization in Charm++ Programs

“…The need for an indexed organization was anticipated and developed in early work by Sanjeev Krishnan and Joshua Yelon [42,48]. These ideas were developed into the notion of a "chare array": an indexed collection of chares [37]. Although they were called "arrays", the index could be quite general, supporting sparse arrays as well as collections indexed by bit-vectors or even strings.…”

“…The system took charge of global location management via a scalable scheme [37], so that it could identify the processor on which the named chare lived, and deliver the message to it.…”

A parallel-object programming model for petaflops machines and blue gene/cyclops

“…Message forwarding may be required after a migration, but the system uses a routing scheme that asymptotically requires only 1 hop (i.e. no indirection) for any kind of repeated communication [9].…”