Accelerating computing with the Cell Broadband Engine processor

Abstract

In this paper, we describe our approach to utilizing the compute power of the Cell Broadband Engine™ (Cell/B.E.)1 processor as an accelerator for computationally intensive portions of high performance computing applications. We call this approach "hybrid programming" because it distributes application execution across heterogeneous processors. IBM developed a hardware implementation and software infrastructure that enables this hybrid computing model as part of the Roadrunner project for Los Alamos National Laboratory (LANL). In the hybrid programming model, a process running on a host processor, such as an × 86-64 architecture processor, creates an accelerator process on an accelerator processor, such as the IBM® PowerXCell™8i2. The PowerXCell8i is a new implementation of the Cell Broadband Engine architecture. The host process then schedules compute intensive operations onto the accelerator process. The host and accelerator process can continue execution concurrently and synchronize when needed to transfer results or schedule new accelerator computation. We describe the Data Communication and Synchronization (DaCS) Library and Accelerated Library Framework (ALF) which are designed to allow applications to create new applications and adapt existing applications to exploit hybrid computing platforms. We also describe our experience in using such frameworks to construct hybrid versions of the familiar Linpack benchmark and an implicit Monte Carlo radiation transport application named Milagro. Performance measurements on prototype hardware are presented that show the performance improvements achieved to date, along with projections of the expected performance on the final Roadrunner system. Copyright 2008 ACM.