Phylogenetics study the evolutionary history of organisms using an iterative process of creating and evaluating phylogenetic trees. This process is very computationally intensive; constructing a large phylogenetic tree requires hundreds to thousands of CPU hours. In this article, we describe an FPGA-based system that can be deployed on AWS EC2 F1 cloud instances to accelerate phylogenetic analyses by boosting performance of the phylogenetic likelihood function, i.e., a widely employed tree-evaluation function that accounts for up to 95% of the overall analysis time. We exploit domain-specific knowledge to reduce the amount of transferred data that limits overall system performance. Our proof-of-concept implementation reveals that the effective accelerator throughput nearly quadruples with optimized data movement, reaching up to 75% of its theoretical peak and nearly 10× faster processing than a CPU using AVX2 extensions.

中文翻译：

---

在云中使用 FPGA 加速系统发育


系统发育学利用创建和评估系统发育树的迭代过程来研究生物体的进化历史。这个过程的计算量非常大；构建大型系统发育树需要数百到数千个 CPU 小时。在本文中，我们描述了一种基于 FPGA 的系统，该系统可以部署在 AWS EC2 F1 云实例上，通过提高系统发生似然函数（即广泛使用的树评估函数，占比高达 95%）的性能来加速系统发生分析。总分析时间。我们利用特定领域的知识来减少限制整体系统性能的传输数据量。我们的概念验证实施表明，通过优化数据移动，有效加速器吞吐量几乎翻了两番，达到理论峰值的 75%，处理速度比使用 AVX2 扩展的 CPU 快近 10 倍。