Pigeon: A Dynamic and Efficient Serverless and FaaS Framework for Private Cloud

EasyChair Preprint 1735

Abstract

Recently, voice-triggered small cloud functions such as Alexa skills [1], and cloud mini programs [2] for IoT
and smartphone, grow exponentially. These new developments also attract organizations to host their own cloud functions or mini programs in private cloud environment and move from traditional Micro-service architecture to Serverless Function-asa-Service (FaaS) architecture. However, current Serverless FaaS
frameworks [3] [4] [5] [6] [7] cannot meet cold start latency, resource efficiency required by short-lived cloud functions and mini programs.

In this paper, we build a new Framework — Pigeon that brings Serverless and FaaS programming paradigm into private cloud to enable enterprises to host these applications. Pigeon creates function-oriented Serverless framework by introducing an independent and finer-grained function-level resource scheduler on top of Kubernetes. A new oversubscription-based static prewarmed container solution is also proposed to effectively reduce function startup latency and increase resource recycling speed for short-lived cloud functions. Empirical results show that Pigeon framework enhances function cold trigger rate by 26% to 80% comparing to AWS Lambda Serverless platform [5]. Comparing to Kubernetes native scheduler based serverless platforms, throughput gets 3 times improvement while handling short-lived functions.

Keyphrases: FaaS, Private Cloud, Serverless, dynamic resource management

@booklet{EasyChair:1735,
  author    = {Wei Ling and Lin Ma and Chen Tian},
  title     = {Pigeon: A Dynamic and Efficient Serverless and FaaS Framework for Private Cloud},
  howpublished = {EasyChair Preprint 1735},
  year      = {EasyChair, 2019}}