Caching techniques in next generation cellular networks

Mahmood, Ahsan

Content caching will be an essential feature in the next generations of cellular
networks. Indeed, a network equipped with caching capabilities allows users to
retrieve content with reduced access delays and consequently reduces the traffic
passing through the network backhaul. However, the deployment of the caching
nodes in the network is hindered by the following two challenges. First, the storage
space of a cache is limited as well as expensive. So, it is not possible to store
in the cache every content that can be possibly requested by the user. This calls
for efficient techniques to determine the contents that must be stored in the cache.
Second, efficient ways are needed to implement and control the caching node. In this
thesis, we investigate caching techniques focussing to address the above-mentioned
challenges, so that the overall system performance is increased.
In order to tackle the challenge of the limited storage capacity, smart proactive
caching strategies are needed. In the context of vehicular users served by edge
nodes, we believe a caching strategy should be adapted to the mobility characteristics
of the cars. In this regard, we propose a scheme called RICH (RoadsIde CacHe),
which optimally caches content at the edge nodes where connected vehicles require
it most. In particular, our scheme is designed to ensure in-order delivery of content
chunks to end users. Unlike blind popularity decisions, the probabilistic caching
used by RICH considers vehicular trajectory predictions as well as content service
time by edge nodes. We evaluate our approach on realistic mobility datasets against
a popularity-based edge approach called POP, and a mobility-aware caching strategy
known as netPredict. In terms of content availability, our RICH edge caching scheme
provides an enhancement of up to 33% and 190% when compared with netPredict
and POP respectively. At the same time, the backhaul penalty bandwidth is reduced
by a factor ranging between 57% and 70%. Caching node is an also a key component in Named Data Networking (NDN) that is an innovative paradigm to provide content based services in future networks. As compared to legacy networks, naming of network packets and in-network caching of content make NDN more feasible for content dissemination. However, the implementation of NDN requires drastic changes to the existing network infrastructure. One feasible approach is to use Software Defined Networking (SDN), according to which the control of the network is delegated to a centralized controller, which configures the forwarding data plane. This approach leads to large signaling overhead as well as large end-to-end (e2e) delays. In order to overcome these issues, in this work, we provide an efficient way to implement and control the NDN node. We propose to enable NDN using a stateful data plane in the SDN network. In particular, we realize the functionality of an NDN node using a stateful SDN switch attached with a local cache for content storage, and use OpenState to implement such an approach. In our solution, no involvement of the controller is required once the OpenState switch has been configured. We benchmark the performance of our solution against the traditional SDN approach considering several relevant metrics. Experimental results highlight the benefits of a stateful approach and of our implementation, which avoids signaling overhead and significantly reduces e2e delays.

Systèmes de Communication
Eurecom Ref:
Politecnico di Torino
See also: