Experiments with the Fresh Breeze tree-based memory model

Abstract

The Fresh Breeze memory model and system architecture is proposed as an approach to achieving significant improvements in massively parallel computation by supporting fine-grain management of memory and processing resources and utilizing a global shared name space for all processors and computation tasks. Memory management and the scheduling of tasks are done by hardware realizations, eliminating nearly all operating system execution cycles for data access, task scheduling and security. In particular, the Fresh Breeze memory model uses trees of fixedsize chunks of memory to represent all data objects, which eliminates data consistency issues and simplifies memory management. Low-cost reference-count garbage collection is used to support modular programming in type-safe programming languages.The main contributions of this paper are: (1) a program exection model for massively parallel computing as the Fresh Breeze application programming interface (API) comprising a radical memory model and a scheme for expressing concurrency; (2) an experimental implementation of the API through simulation using the FAST simulator of the IBM Cyclops 64 many-core chip; (3) -mail: meng@capsl.udel.edu that demonstrate that (a) fine-grain hardware-implemented resource management mechanisms can support massive parallelism and high processor utilization through the latencyhiding properties of multi-tasking; and (b) hardware implementation of a work stealing scheme incorporated in our simulation can effectively distribute tasks over the processors of a many-core parallel computer.

Massively parallel breadth first search using a tree-structured memory model

“…Earlier simulations demonstrating the merit of the Fresh Breeze PXM for standard linear algebra computations, dot product, matrix multiply and the Fast Fourier Transform, have been reported in recent publications [10,11], showing effectiveness in exploiting fine-grain concurrency in these kernels. The work reported here adds to our confidence that the Fresh Breeze PXM is a worthy candidate for guiding the architecture of future massively parallel computer systems.…”

“…Programs written using a library interface to an implementation of the Fresh Breeze PXM have been developed for the linear algebra kernels, dot product, matrix multiply, and the Fast Fourier Transform. Our simulation tool has shown that the use of the Fresh Breeze memory model in these kernel algorithms achieves full utilization of 40 processing cores even for modest problem sizes [10,11]. Section 6 of this paper reports on simulation experiments using the breadth-first search algorithm discussed in Section 4.…”

On the Feasibility of a Codelet Based Multi-core Operating System

“…For use in massively parallel high performance systems, these implementations have unacceptable complexity and energy consumption. Fortunately, this need not be the case: we have demonstrated a memory model implementation, using trees of fixed size memory chunks, that provides an efficient global virtual memory for all executing tasks in a multi-core system [8].…”

“…A way out of the garbage collection dilemma is to adopt a program execution model in which data blocks, once defined by execution of a codelet, are never modified. This write-once policy guarantees that cycles of references can never be created, and therefore the reference count method of garbage collection [8] may be used. Moreover, reference count garbage collection is readily implemented in hardware as a distributed activity, concurrent with user task execution.…”

“…Replacing these functions of the runtime software with efficient hardware mechanisms has the potential to yield major improvements in performance and energy efficiency. The benefits of hardware implementation of resource management functions has been demonstrated in simulations of the Fresh Breeze system architecture [8].…”