The radio access point in an LTE (Long Term Evolution) mobile network, called an e-NodeB or Radio Base Station (RBS) is very much a high performance, Embedded platform. The system is a heterogeneous platform consisting of general-purpose CPUs, high performance DSPs, and hardware accelerators. From a software parallelism point of view it covers highly complex data parallelism in the user plane domain as well as massive task based concurrency in the control plane domain. It is common to use many different programming models and even virtualisation scenarios in engineering the system and software in order to satisfy the diverse properties and characteristics of the system.
As we look towards the future with a vastly increased number of connected devices and increasingly dense and mixed cell and radio configurations, the energy-performance equation becomes a critical factor in constructing high-performing, energy-efficient and cost effective system solutions. The ability to construct abstract models and reason about this balance would be a crucial enabling factor.
In Adept we propose to apply the developed technology to model parts of our e-NodeB control/user plane access software on different hardware possibly with “big.LITTLE” configurations and using different parallel programming models. Using this technology we expect to be able to create trade-off scenarios regarding optimum power-performance partitioning of parallel software and algorithms across heterogeneous and “big.LITTLE” platforms.