TY - GEN
T1 - DirectFaaS
T2 - 33rd ACM Web Conference, WWW 2024
AU - Zeng, Qingyang
AU - Hou, Kaiyu
AU - Leng, Xue
AU - Chen, Yan
N1 - Publisher Copyright:
© 2024 ACM.
PY - 2024/5/13
Y1 - 2024/5/13
N2 - Serverless computing, also known as Function-as-a-Service (FaaS), triggers web applications in the form of function chains. It uses a central orchestrator to route all requests from end-users and internal functions. Such architecture simplifies application deployment for developers. However, the convenient centralized network architecture compromises the efficiency of function chain communications. Specifically, (i) a centralized API gateway assists in routing requests between functions. This indirect routing scheme raises invocation latency. (ii) The control flow for invoking functions and the data flow for passing function data packets are both forwarded by the API gateway. This results in the API gateway consuming a significant amount of resources. (iii) All data packets of internal function communications go through the same API gateway. This expands the additional attack surface in multi-tenant scenarios. In this paper, we propose DirectFaaS, a clean-slate network architecture to improve the function chain communication performance. By separating coupled control flow and data flow, DirectFaaS releases the API gateway from heavy traffic forwarding, reducing its resource consumption. For this goal, DirectFaaS exploits the network control capabilities of Software-Defined Networking (SDN) to establish direct data forwarding channels to accelerate function chain invocations. In addition, the data flow constrained by fine-grained network policies consolidates multi-tenant traffic security. We implement the DirectFaaS prototype on the popular OpenFaaS platform. Evaluations under real-world serverless applications show that DirectFaaS achieves a reduction in application execution time by up to 30.9% and CPU consumption by up to 30.1% compared to the current architecture.
AB - Serverless computing, also known as Function-as-a-Service (FaaS), triggers web applications in the form of function chains. It uses a central orchestrator to route all requests from end-users and internal functions. Such architecture simplifies application deployment for developers. However, the convenient centralized network architecture compromises the efficiency of function chain communications. Specifically, (i) a centralized API gateway assists in routing requests between functions. This indirect routing scheme raises invocation latency. (ii) The control flow for invoking functions and the data flow for passing function data packets are both forwarded by the API gateway. This results in the API gateway consuming a significant amount of resources. (iii) All data packets of internal function communications go through the same API gateway. This expands the additional attack surface in multi-tenant scenarios. In this paper, we propose DirectFaaS, a clean-slate network architecture to improve the function chain communication performance. By separating coupled control flow and data flow, DirectFaaS releases the API gateway from heavy traffic forwarding, reducing its resource consumption. For this goal, DirectFaaS exploits the network control capabilities of Software-Defined Networking (SDN) to establish direct data forwarding channels to accelerate function chain invocations. In addition, the data flow constrained by fine-grained network policies consolidates multi-tenant traffic security. We implement the DirectFaaS prototype on the popular OpenFaaS platform. Evaluations under real-world serverless applications show that DirectFaaS achieves a reduction in application execution time by up to 30.9% and CPU consumption by up to 30.1% compared to the current architecture.
KW - sdn
KW - serverless computing
KW - serverless function chain
KW - serverless networking
UR - http://www.scopus.com/inward/record.url?scp=85194053259&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85194053259&partnerID=8YFLogxK
U2 - 10.1145/3589334.3645333
DO - 10.1145/3589334.3645333
M3 - Conference contribution
AN - SCOPUS:85194053259
T3 - WWW 2024 - Proceedings of the ACM Web Conference
SP - 2759
EP - 2767
BT - WWW 2024 - Proceedings of the ACM Web Conference
PB - Association for Computing Machinery, Inc
Y2 - 13 May 2024 through 17 May 2024
ER -