The Abaco Platform: A Performance and Scalability Study on the Jetstream Cloud
The Abaco Platform: A Performance and Scalability Study on the Jetstream Cloud
复制标题
Abaco 平台：Jetstream 云的性能和可扩展性研究
DOI：
10.1007/978-3-030-69984-0_77
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
Nguyen, Kreshel
中科院分区：
文献类型：
--
作者：
Garcia, Christian;Stubbs, Joe;Looney, Julia;Jamthe, Anagha;Packard, Mike;Nguyen, Kreshel
Abaco是一个开源的分布式云计算平台，基于并行计算的Actor模型和Linux容器，由美国国家科学基金会资助，托管于德克萨斯高级计算中心。Abaco最近实现了一个自动缩放功能，允许根据参与者的邮箱队列长度自动缩放参与者的工作池。在本文中，我们解决了几个与Abaco平台与手动和自动缩放功能的性能相关的研究问题。通过系统地研究Abaco在各种场景下的总FLOPS和哈希吞吐量来测试性能和稳定性。通过测试，我们确定Abaco可以正确地缩放到100 Jetstream“m1”。medium”实例，并实现超过19 TFLOPS。
Abaco is an open-source, distributed cloud-computing platform based on the Actor Model of Concurrent Computation and Linux containers funded by the National Science Foundation and hosted at the Texas Advanced Computing Center. Abaco recently implemented an autoscaler feature that allows for automatic scaling of an actor’s worker pool based on an actor’s mailbox queue length. In this paper, we address several research questions related to the performance of the Abaco platform with manual and autoscaler functionality. Performance and stability are tested by systematically studying the aggregate FLOPS and hashrate throughput of Abaco in various scenarios. From testing we establish that Abaco correctly scales to 100 Jetstream “m1.medium” instances and achieves over 19 TFLOPS.