In the race to keep mobile performance in line with user demands, carriers around the world are spending big on splitting their cellular networks into ever-smaller chunks. New research suggests that adding some spinning rust to base stations could help improve efficiency, at lower cost.
The modelling, carried out by Ejder Baştuğ and Mérouane Debbah of SUPÉLEC, Gif-sur-Yvette, France and Mehdi Bennis of the Centre for Wireless Communications at the University of Oulu in Finland, looks at what in-network caching could do for cellular network performance.
Their base assumption is that to cope with the capacity crunch, carriers and service providers are inevitably going work with short-range coverage models – “het-nets”, WiFi handoff, nanocells, and small cell networks (SCNs).
The problem with such deployments is that they put increasing demands on backhaul networks: either the carrier has to roll out more backhaul to cope with the extra base stations, or split the backhaul across a larger number of base stations.
The practicality of expanding the backhaul depends on the infrastructure that's available: base stations served by fibre don't pose too much of a problem, but many base stations are served by microwave or copper tails.
Hence the interest in caching, since it can reduce the demand on the backhaul network. It's a model familiar to anybody with knowledge of the operation of Internet services, whether it's a Squid proxy in an ISP or the large-scale caching built into content distribution networks (CDNs).
The SCN, however, is a constrained environment: any caching deployed to a miniature base station has to fit within much tighter storage constraints. So the SUPÉLEC / Oulu paper sets out mathematical models for getting caching to work in such a tight environment.
As they say, the model takes into account parameters like signal-to-interference-plus-noise ratio (SINR), base station location, target file bitrate, storage size and the popularity distribution of individual files. The inclusion of SINR is interesting: as they note, “this formulation is the first attempt” to couple cache design with the physical layer.
They argue that the downlink performance matters in designing a caching model for wireless because of the way users interfere with each other. In a fixed broadband network (with the exception of HFC), serious contention starts at the first backhaul link – for example, the backhaul from the DSLAM upstream to the ISP*.
In the wireless network, the number of users served by a base station has a direct and immediate impact on everyone, so the authors have produced mathematical models that take this into account in cache design.
At this point, however, this hack reaches the limit of his ability to decode the maths:
However, the trade-off the paper presents is one that will be of interest to carriers: although the best way to improve end user performance is to build out more base stations, with the right cache design, adding storage to the base station can mitigate the need to build out extra backhaul.
The idea of implementing such caches is not entirely far-fetched. Synology recently released a semi-rugged NAS that looks like it could handle some of the job. Cellular base stations like Cisco's 7000-series Small Cells include ethernet, so could hook up to a NAS on a small local LAN. ®
*Bootnote: Yes: in a copper cable bundle, users appear to each other as interference. However, DSL deployment rules are designed to minimise the impact of this interference on individual links. If the paper has put forward a useful scheme for incorporating the physical layer SINR into cache design, it might also be useful in DSL networks. ®