HCloud: A Serverless Platform for JointCloud Computing

Jinfeng Liu,Zeyu Mi,Zheng Huang,Zhichao Hua,Yubin Xia

DOI: https://doi.org/10.1109/jcc49151.2020.00022

2020-01-01

Abstract:Most of today's cloud providers, including Amazon, Google and Microsoft, usually offer computing services with the abstraction of virtual machines (VMs), which are also known as instances. Even though these public clouds share similar infrastructures, they provide different service quality standards and price models for users, which fluctuate according to the dynamic loads. The JointCloud computing model empowers the cooperation among multiple public clouds, which can provide the users better service quality with reasonable prices. However, it is difficult to dynamically cooperate among clouds based on the traditional VM abstraction. Any computation migration among clouds will trigger costly VM states transfer, which makes it extremely hard for users to adjust their deployment models swiftly. Fortunately, serverless computing model is getting popular recently and has been supported by major cloud vendors. This model allows users to upload their computing tasks to the cloud in the unit of function. Users do not need to consider all of the tedious works like virtual server maintenance; instead, the cloud automatically instantiates function workers to handle incoming requests on the fly. In this paper, we propose a new JointCloud platform called HCloud, which efficiently manages computing resources of multiple clouds while offering the server-less model to the users. The fine-grained granularity of function enables HCloud to flexibly migrate workloads among clouds, thus brings better service qualities and lower prices at the same time. For example, HCloud can reduce the cost by requesting resource from a cheaper cloud for latency-insensitive jobs while routing the requests of latency-sensitive ones to the nearest and performant clouds. We have implemented a prototype of HCloud and evaluated it by simulating multiple cloud providers. The evaluation results show that HCloud can greatly improve the performance of serverless workloads with small costs.