Real-time Serverless: Cloud Resource Management for Bursty, Real-time Workloads

Hai Nguyen; Chaojie Zhang; Zhujun Xiao; Andrew Chien. 16 July, 2019.
Communicated by Andrew Chien.


Today’s cloud resource offerings provide no guarantees for resource allocation, so bursty application must reserve, and pay for resources they do not use - to achieve real-time guarantees. We propose a new type of cloud resource, Real- time Serverless (RTS) with a new service-level objective – guaranteed allocation rate. This guarantee enables timely re- source allocation, enabling applications to achieve real-time performance efficiently. With a simple burst model, we study real-time serverless analytically, exploring their effect on ap- plication quality, guarantees, and cost. Next, we simulate statistically varying bursts and higher loads (multi-application), to study the impact of real-time serverless.

In both analytic and simulation studies, adding guaran- teed allocation rate enables bursty, real-time applications to achieve guaranteed high quality cost-effectively. Specifically, for a desired application quality, the required allocation rates can be determined. In addition, for duty factors from 0.025 to 0.25, the value of real-time serverless to the application is > 4x than traditional. Further results show that multiple applications can share real-time serverless efficiently, supporting duty factor increases of 25x with only a 1.6x increase in allocation rate (provider resource cost). Finally, we present a case study of a traffic monitoring application. Despite more complex burst statistics, our results show major benefits for cost and application quality. Application benefit makes real-time serverless worth nearly 16x virtual machine resources for delivering application value.

Original Document

The original document is available in PDF (uploaded 16 July, 2019 by Andrew Chien).