Serverless linear algebra
Serverless linear algebra
复制标题
DOI：
10.1145/3419111.3421287
复制
发表时间：
2020-10
期刊：
影响因子：
--
通讯作者：
Vaishaal Shankar;K. Krauth;Kailas Vodrahalli;Qifan Pu;B. Recht;I. Stoica;Jonathan Ragan-Kelley;Eric Jonas;S. Venkataraman
中科院分区：
文献类型：
--
作者：
Vaishaal Shankar;K. Krauth;Kailas Vodrahalli;Qifan Pu;B. Recht;I. Stoica;Jonathan Ragan-Kelley;Eric Jonas;S. Venkataraman
数据中心分解为数据中心操作员和应用程序设计人员提供了许多好处。然而，从以服务器为中心的模型切换到分解模型需要开发新的编程抽象，这些抽象可以实现高性能，同时受益于更大的弹性。为了探索数据中心分解的局限性，我们研究了一个应用领域，它几乎最大限度地受益于当前以服务器为中心的数据中心：密集线性代数。我们构建了NumPyWren，一个建立在分散的无服务器编程模型上的线性代数系统，以及LAmbdaPACK，一种专为高度并行线性代数算法的无服务器执行而设计的同伴特定于域的语言。我们表明，对于一些线性代数算法，如矩阵乘法，奇异值分解，Cholesky分解和QR分解，NumPyWren的性能（完成时间）是优化的以服务器为中心的MPI实现的2倍之内，并具有高达15%的计算效率（总CPU小时），同时提供容错。
Datacenter disaggregation provides numerous benefits to both the datacenter operator and the application designer. However switching from the server-centric model to a disaggregated model requires developing new programming abstractions that can achieve high performance while benefiting from the greater elasticity. To explore the limits of datacenter disaggregation, we study an application area that near-maximally benefits from current server-centric datacenters: dense linear algebra. We build NumPyWren, a system for linear algebra built on a disaggregated serverless programming model, and LAmbdaPACK, a companion domain-specific language designed for serverless execution of highly parallel linear algebra algorithms. We show that, for a number of linear algebra algorithms such as matrix multiply, singular value decomposition, Cholesky decomposition, and QR decomposition, NumPyWren's performance (completion time) is within a factor of 2 of optimized server-centric MPI implementations, and has up to 15% greater compute efficiency (total CPU-hours), while providing fault tolerance.