CLFBP : 手機及Linux程式之異質性伺服器雲端加速平台

Abstract

智慧型手機改變了人類的生活，而如何讓手機的應用程式跑得更快更順暢，實是近年來熱門的研究題材。手機的運算有先天的硬體限制。為了加速手機程式，Migration是一個常用的技術。 我開發了一套嶄新的運算平台CLFBP,結合兩大概念OpenCL以及FBP，透過OpenCL可以利用異質性平台的優勢，而透過FBP可以做程式的動態搬移。因此CLFBP可以同時解決Migration會遇到的四大難題，包括程式切割、片段溝通、動態決策以及善用異質性平台優勢；此外，CLFBP還提供了一套方便好用的API，讓程式開發者能夠快速的開發出自己的FBP程式並且享有CLFBP平台的功能。

Smartphone has changed human life, and making smartphone applications smoother and more efficient has become a hot research topic in recent years. Both of the computing capability and the power consumption of smartphones are limited due to the small form factors. Task offloading is a common technology to speed up the performance of smartphone applications. Based on the concept of flow-based programming (FBP), we propose a framework, called CLFBP, which combines OpenCL into FBP to provide two benefits: Using OpenCL, a CLFBP program can leverage the power of heterogeneous servers; Using FBP, smartphone applications can migrate part of its workload to the cloud. CLFBP addresses four challenges for migrating workload in today's mobile-cloud environment including tasks partition, facilitating communication between tasks, making dynamic offload decisions, and leveraging the power of heterogeneous servers. In addition, CLFBP provides application programming interfaces for both coarse-grain task partitioning and fine-grain data parallel programming, which allows programmers to re-factor applications conveniently.