Home Technical objectives
Technical objectives

The major idea underlying femto-clouding is to merge cloud computing with femtocell networking, thus enabling a capillary distribution of the cloud computing capabilities closer to a potentially huge number of mobile users. The main challenge is how to distribute the computation/communication capabilities between mobile handsets and the cloud efficiently, in order to deliver services with provable QoS, in terms of latency, service continuity, etc. To tackle this global task, it is necessary to solve a series of sub-tasks pertaining to advanced physical layer and radio resource management algorithms, virtualisation techniques, coordination between cloud service provisioning and femtocell networking, as detailed in the following sections.

Enhancements in advanced channel aware PHY and RRM techniques

A massive deployment of user-operated femto-access points induces an adverse interference scenario that calls for the use of advanced radio resource management supported by LTE-A and its evolutions. Different levels of coordination/cooperation among HeNBs are key to enhance system capacity and keep interference to an adequate level. To that end, Coordinated Multipoint Communication (CoMP) and Multipoint-to-Multipoint (MP2MP) communication techniques are receiving increased attention as PHY layer technologies providing significant capacity gains. Interference Alignment (IA) techniques have also been shown to be theoretically appealing as a way to handle interference channels efficiently. One of the goals of this project is to extend CoMP and MP2MP to the heterogeneous network scenario composed of the advanced HeNBs foreseen in TROPIC and MeNBs, with the possible inclusion of low power base stations to cover small cells.

One of the specific goals is to make MP2MP and IA applicable within an LTE-A framework, under computation/communication constraints induced by the hardware available at the HeNBs and the quality of the backhaul links. Some of the most significant steps forward foreseen in this context are listed below:

  • Multipoint-to-Multipoint communication represents the most general setting to analyze collaborative communication schemes. In conventional CoMP techniques, collaboration is foreseen among base stations to mainly improve efficiency of cell edge users. In TROPIC, we will extend CoMP to the femtocell scenario, where different levels of cooperation between HeNBs and between HeNBs and MeNB are key elements to enhance system capacity and keep interference to an adequate level. Based on today's technology, one critical aspect is that the ADSL link between HeNBs is not considered to be secure. To ensure secure communications between HeNBs and between HeNB and MeNB, a better approach is to consider over-the-air (OTA) communications, not yet supported by LTE. OTA communications may be also useful to synchronize HeNB, without being prone to unpredictable latencies of the ADSL links. Within this general MP2MP set-up, conventional broadcast, multiple access channels or interference channel are only particular limiting cases. In all configurations, interference alignment and null-steering have been shown to be the right tools for high performance at high SNR, but fail short at medium-to-low SNR. TROPIC proposes to leverage on distributed MMSE criteria to simplify the implementation of precoders/decoders and adapt those strategies to LTE framework.
  • Energy-efficient computation offloading: Since energy consumption at the mobile handset is one of the major limitations to the development of more and more sophisticated applications for mobile handset, TROPIC will analyze mechanisms to offload part of the computations at the mobile site in order to find the optimal balance between the energy spent for processing and the energy spent for communication, under delay constraints imposed by the application and by the channel state.
  • Decentralized resource allocation: In view of a potential massive deployment of HeNBs, it will be fundamental to devise decentralized mechanisms for radio resource management and distributed computation. Game-theory (GT) has been proved to be a powerful tool to devise decentralized radio resource allocation techniques. In particular, game-theoretic algorithms for decentralized resource allocations in femtocell networks have already been devised and analyzed in the FP7 FREEDOM project, whose partners are also composing the consortium of TROPIC. Building on the experience gained with FREEDOM, one of the main innovations in TROPIC will be that the utility function of each player (node) will now incorporate not only communication relevant metrics, such as energy or spectral efficiency, but also computational metrics, which take into the requirements of the application running in the mobile handset and the current level of battery. The result will be a dynamic allocation of radio resources and computational tasks. From this more general perspective, latency will be a critical parameter, especially for delay sensitive applications, like voice or video streaming.
  • Collaborative sensing: One of the key elements to allocate radio resources efficiently requires learning from the environment and adaptation. However, learning in a single location is subject to local fading or shadowing phenomena. To counteract this inconvenient, collaborative sensing techniques will be explored within the project exploiting the exchange of information between sensing devices installed in HeNBs, taking into account the limitations resulting from exchanging data over a best effort channel. Since in a realistic environment, detection errors and estimation inaccuracies are going to affect the system efficiency, we will analyze a joint optimization of detection thresholds and RRM, in order to optimize relevant performance metrics, such as transmit information rate or transmit power, under constraints on the maximum interference generated towards other users.

Device to Device (D2D) communications. Device to device communication is a new mode of data transfer between mobile terminals that are in the same radio range. This mode is controlled by the network (HeNB or macro base-station), however the exchanged data are not transferred through the HeNBs. TROPIC will investigate appropriate PHY level interference management schemes, resource allocation strategies and signalling procedures in order to enable D2D communications. The main objective of these strategies is to allow the reuse of the radio resources that are dedicated to D2D links for other communications within the femto-cell. Such feature will provide a better usage of the spectrum within femtocells. Also it alleviates the computation load on the HeNBs which does not need to decode all the messages that are transferred with this mode of communication.

Enhancements in coordination procedures

Radio resource management is a critical aspect in femto-cell networks. In fact since HeNBs are deployed by home users, prior planning is not possible and interference level will depend on the radio resource allocation mechanisms that are used by each HeNB. The developed strategies have to be feasible with the available computing resources and also have to be scalable when taking into account the potential number of deployed HeNBs.

Because of the dynamic and unpredictable nature of the radio environment, distributed schemes offer a competitive approach to meet these requirements. TROPIC proposes to investigate distributed radio resource allocation schemes by leveraging the new features that are developed in the HeNBs namely: sensing and the cloud computing infrastructure. Tools from Game Theory, stochastic and robust optimization will be used to tackle radio resource problematic in this new context. Such tools have shown their ability to provide efficient
methods to these problems and the consortium will build on deep knowledge that has been acquired within previous projects (FREEDOM being one of them). The proposed investigations will focus on autonomous and distributed approaches by taking into account realistic knowledge provided by sensing mechanisms and by leveraging the availability of cloud computing infrastructure.

Another challenge is to ensure that users will remain connected to the same HeNB during data processing. Otherwise, it could be quite complicated to deliver processed data back to the user. Hence, we will analyze and study possible approaches how to prevent handovers from the HeNB that performs computation for the UE. The promising method is to set transmission power of all involved HeNBs adaptively in such manner that the user will not perform handover to an adjacent cell. Another way is to smartly adjust hysteresis
margin and thus to hold user at the same HeNB.

New paradigms for virtualisation and management

In computer science, virtualisation refers to the creation of a virtual machine that behaves like a real computer with an operating system (a sort of container). The services or software applications running on these Virtual Machines (VM) are made independent of the real hardware resources or computer. Most common activities in the current virtualization area deal with server virtualization mainly in clusters or server farms of datacenters.

The hypervisor is one of many existing hardware virtualization techniques allowing multiple operating systems, called guests, to run concurrently on a computer. Hypervisors are installed on the computers whose only task is to run guest operating systems. Some well-known solutions are Xen, KVM, etc. The control of the VMs requires a Virtual Infrastructure Manager (VIM) -Citrix, VMware, Open Nebula, Eucalyptus, OpenStack Compute, etc.- in charge of the VM lifecycle and implementing functionalities like elasticity. This describes the interface (API) provided by the specific cloud-computing functionality infrastructure as a service (IaaS). The IaaS consists in the delivery of a raw computer infrastructure by means of a simple API.

In TROPIC, we will study the characterization of particular hardware resources that are the femto stations. The base station is specific purpose hardware with limited access to the CPU, disk, memory and the network bandwidth of the radio-links. We will analyse the suitable virtualisation technique (hypervisor) for the HeNBs so that it can be manage in a cloud-like approach by a VIM. With regard to the API we shall consider the adaptation of existing specifications like the Open Cloud Computing Interface (OCCI) [occi-wg.org]. OCCI aims to specify a remote management protocol for cloud services. Implementations exist or are in development for a number of cloud platforms and tools. While OCCI provides a relatively low level interface for managing cloud services, federated cloud models require a higher level mechanism to specify the deployment of complete cloud applications rather than individual resources. This functionality is largely provided by the Open Virtualization Format (OVF); however, as OVF itself does not include concepts such as elasticity and service levels and needs to be extended by a policy-based approach.

It will be necessary the inclusion of these Cloud concepts in the coordination of Femto networks. For instance, TROPIC envisages the addition of a new functional entity, named Femto-Cloud Manager (FCM), within the fixed operator network, as part of the MeNB or the HeNB GW (see also Fig. 11, pg. 13). The Femto-Cloud Manager is seen by TROPIC as an extended VM Manager; its main functionalities are (but are not limited to):

  • Virtual Machine placement across multiple hypervisors3;
  • modify the VM resources dynamically (disk size, memory space, networking, ...);
  • VM monitoring at runtime; scheduling capabilities based on policies;
  • user management and VM management across multiple zones;
  • self-environment. (e.g. self-service, self-managing, self-healing, self-reporting, self-provisioning, to be defined in the scope of Femto Cloud case);
  • fault tolerance system at infrastructure level (usually composed by a backup system and a recovery
    system using snapshots);
  • elastic capabilities - Bursting. When the internal resources of an organization are not able to address a spike of demand then these systems are able to burst to a third-party provider.

In turn, the management of these particular resources (femto stations) will imply to study and incorporate statistical models at the management layer in order to consider the dynamicity of the scenario like the availability of the HeNB is not always guarantee since it is distributed across homes. Another novel interesting topic would be the creation of federated clouds matching the femto network topology.

Advances in the system level integration

Distributed Computing on a femto-cloud system. Femto-clouding has the potential to efficiently support mobile applications that are out of reach for modern handsets. Applications on mobile devices are constrained by limited resources as low frequency CPU, small memory and limited battery power. In TROPIC the mobile device acts as a possible thin device with most of the computation/storage occurring at the edge of the femto-cloud. We believe that such a solution will make possible to fully exploit the immersive and rich experience made possible by today portable device without incurring to the (severe) limitations imposed by their constrained resources.

The femto-cloud layer provides cheap, low latency, and readily available computation and storage resources that can be used to off-load real time applications from the handset in such a way to achieve much larger battery lifespan and to enable applications that cannot execute on the handset alone. The development of this system opens up several exciting research challenges in the areas of distributed systems and networking.

Hand-set/femto-cloud system. To fully exploit the femto-cloud concept we need to consider a programming style which allows for distributed implementation on the femto-cloud system of a generic application. While some work has been done in projects like MAUI e J-Orchestra [MAUI10, J-Orch09], additional and totally new notions have to be developed to describe the distributed application in terms of computation, storage, energy, and real-time requirements:

  • Hand-set/femto-cloud system interface: We need a clean and simple interface that the handset can use to off-load computation and/or storage to the femto-cloud system. The interface is used to define the QoS requirements of the application that is being offloaded---real-time, storage, computation, energy, and communication requirements.
  • Optimization of load between the hand-set and the femto-cloud layer.
  • Cloud/femto-cloud system interface: We will investigate mechanisms and rules to decide when and how to move computation and storage between the cloud and the femto-cloud layers. For instance we expect that offloading non real-time tasks to the cloud system in case of overloading of the femto cloud system will be a mean to increase the efficiency of the overall system.
  • Mobility management: While we can expect that handsets have low mobility in the scenarios of interest (home, workplace, etc.), moving cloud computing towards the handsets introduces the challenge of handling mobility by keeping the prescribed quality of service in terms of delay and overall efficiency.

Resource allocation within the femto-cloud layer. A thorough understanding of the trade-offs associated to static and dynamic HeNB resource allocation to users will be achieved. We expect that simple solutions will be the most effective under light load but that optimization techniques will be needed as the amount of tasks offloaded to the femto cloud grows. Our approach will be that of first developing analytical frameworks (analytical models, bounds) with the objective to study optimum resource allocation in a dynamic femto cloud environment. Based on such models we will then investigate localized and distributed heuristics for HeNB resource allocation in an heterogeneous setting in terms of task QoS requirements, demand, importance, and available resources at the HeNB. Lightweight schemes which require limited exchange of information among HeNBs while being able to achieve near optimum performance will be designed.

Cross-layer optimization. If we look at the resource allocation and optimization from a broader perspective, we can regard the femto-cloud and the femto-cells as a coupled system whereby the femto cells provide the network resources to support mobile-to-mobile and mobile-to-cloud communications. Hence, rather than considering and optimizing the two (sub)-systems independently, we will explore the possibility of jointly optimizing the system as a whole: radio resource allocation and interference management optimization will be carried out by taking into account applications QoS requirements and user utility and - at the same time cloud resource allocation and computation/storage off-loading will take into account the requirements and the utility the femto-cell operator. As for RRM techniques, we will exploit game theory and/or distributed optimization techniques to device both competitive and cooperative solutions depending on the scenario.

Demonstration platform

In TROPIC, the activity addressed to a demonstration platform has a twofold aspect: On the one hand, the demonstration platform is a necessary internal verification tool to implement, verify and test the achievements of the project; on the other hand, it provides feedbacks to the other RTD activities. Moreover, the industrial partners will invite interested manufacturers, operators and providers to give their comments on the results provided by the activity, along with the overall feasibility. In this perspective the demonstration platform will be a way to reinforce the impact of TROPIC, by showing the achievable results of the project to a wider audience and further enriching the validation activities.

The algorithms and protocols developed in WP3, WP4 and WP5 will be implemented and verified through a system level simulator and a hardware prototype, both developed in WP6. The former gives sufficient evidence of the performance of each element of the system, but lacks tangible evidence; the latter will provide such tangible evidence, but will focus on a simplified case implementing a limited, but meaningful selection of the scenarios, models, and architectural design.


Latest News