Scaling persistent connections for cloud services

The unprecedented growth of cloud applications for connected devices is being hampered by scalability and associated costs for maintaining persistent connectivity between the cloud services and massive numbers of clients. In this paper we present the design and implementation of Connection Parking, a scalable and cost-effective cloud service delivery mechanism to millions of cloud-connected devices. In particular we focus on push delivery mechanisms where it is not possible for the cloud service to initiate connection establishment with the device. Our approach reduces the state resource requirements on the cloud servers without significant adverse impact on service delivery delay. We have prototyped and evaluated our solution with an operational stack of a cloud delivery service that has been designed, commercially deployed and is in production to support millions of connected devices. Our initial results demonstrate that Connection Parking can improve the scalability of cloud services with persistent connections by 3X under different operational scenarios. Our approach can be deployed in a transparent manner and does not require any modifications to either the cloud stack or the client device.