.png){kind=icon}
.png){kind=icon}
Return
Efficient and Flexible Component Placement for Serverless Computing
.png){kind=icon}
PRE
AI
DOI：10.1109/JSYST.2024.3381590.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Nowadays, serverless computing has been widely employed and viewed as the new paradigm of cloud computing. Technically, serverless applications are made up of function components, which are packaged as specific layered files named container images. In production, different components are designed to partially share layers, and during the deployment, the hosting servers have to download the missing layers first, which might dominate the application startup delay. In this article, we look into optimizing the deployment of serverless applications under the operational goals of energy saving and load balance, by exploring the reusability among involved container images to conduct content-aware component placements explicitly. We find that the two involved optimization problems can be formulated as multi-objective (mixed-)integer linear programs, and prove that their common building block of minimizing the weighted sum of deployment cost for a given set of serverless components is non-deterministic polynomial (NP)-hard. To be practical, we develop an efficient yet flexible heuristic solution named best fit greedy placement (BFGP), which involves three variants BFGP-Full, BFGP-ES, and BFGP-LB for the problem. Performance studies show that BFGP is effective, expressive, and efficient. It not only achieves near-optimal placement very efficiently but also supports high-level operational policies, such as energy saving and load balance.
Keywords:
Costs
Containers
Cloud computing
Energy conservation
Serverless computing
Delays
Production
Cloud
energy saving
load balance
serverless
service placement
Journal
I
IF:
2.4
Papers:
4.5K
Citations:
387