Dynamic Resource Management of Cloud NativeControl Units in 5G Radio Access Networks
2018
The ambitious goals of Fifth Generation (5G) mobile networks for higher system capacity, massive
number of devices and flexibility in operations demand the network architectures to be much more
flexible, efficient and autonomous. The design of Radio Access Network (RAN) is undergoing architectural
transformations to increase the flexibility of deployment and programmability by leveraging
Network Functions Virtualization (NFV), Software Defined Networking (SDN), and Cloud Computing.
Hence, efficient resource management strategies with enhanced service quality play a vital
role in realizing true benefits of 5G. Due to spatio-temporal variation of end users data traffic requirements,
traffic load experienced by base stations present at different cell sites exhibit highly
dynamic behavior in traditional cellular systems. This non-uniform and dynamic traffic load may
lead to under utilization of the base station computing resources at cell sites. Cloud Radio Access
Network (C-RAN) is an innovative architecture which addresses this issue and keeps the Total Cost
of Ownership (TCO) under safe limit for cellular operators.
In C-RAN, the baseband processing units, abbreviated as BBUs or Control Units (CUs) are
segregated from cell sites and are pooled in a cloud data center thereby facilitating shared access
for a set of Remote Radio Units (RRUs) present at cell sites. Each CU serves a designated RRU
(1:1 mapping) over a transport network called fronthaul adhered to strict latency and bandwidth
requirements. In this work, various dynamic resource management strategies are presented for 5G
Cloud Radio Access network (C-RAN) considering spatio-temporal traffic heterogeneity exhibited
at Remote Radio Units (RRUs). In C-RAN, the baseband RAN functions (Control Units) are
implemented as virtualized network functions (VNFs) whose life-cycle can be efficiently managed by
cloud platform management and service orchestrator (MANO). The allocation of virtualized instance
of CUs on compute servers is formulated as a classical bin packing problem, where each server is
treated as a bin with finite resources (in terms of Floating Point Operations Per Second (FLOPS))
and each CU is considered as an item to be packed into the bins of equal capacity. An Integer Linear
Programming (ILP) model to minimize the total number of active servers in the data center and
to improve the QoS of ends users is formulated. The ILP model is solved with GAMS tool using
standard solver and the run time performance is analyzed. To alleviate the computational heaviness
of the ILP model for larger input size, two classes of greedy heuristic algorithms : (1) Relocation-
Oblivious algorithm and (2) Relocation-Aware algorithm are devised to solve the problem. Also, by
using analytical approaches of two-sided matching theory and college admission game, a modified
version of classical Deferred Acceptance Algorithm (DAA) algorithm is presented as an efficient
method to solve the mapping of variable size CUs to the compute servers.
A real-world testbed has been setup using an open-source implementation of 3GPP compliant
cellular stack, OpenAirInterface (OAI) along with USRP-B210 SDR and commercial general purpose
processor (GPP) servers. The deployment of the centralized CU on a remote server, and the RRUs,
connected over Ethernet fronthaul is demonstrated. Further, this testbed illustrates the flexibility
in deploying several cellular radio access network protocol split architectures using OAI.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
0
References
0
Citations
NaN
KQI