Elias A. Doumith

Resource Provisioning for Enriched Services in Cloud Environment

Cloud services are based on the provisioning of computing, storage, and networking resources in order to satisfy requests generated by remote end-users. High speed Internet access and multi-core Virtual Machines (VMs) enable today the provisioning of diversified and enriched types of services in Cloud environment. In this paper, we consider several types of basic services and show how their orchestration may lead to the provisioning of more sophisticated services. For this purpose, we define four types of requests that cover the wide spectrum of possible services. We then formulate the resource provisioning problem as a Mixed Integer Linear Program (MILP). We assume that the underlying infrastructure is based on a set of end-to-end connections with guaranteed sustainable bandwidth such as Carrier-Grade Ethernet (CGE) circuits. We investigate the impact of two innovative services on resource allocation carried out by a Cloud Service Provider (CSP). These services correspond to distributed data storage and to multicast data transfer. For the former service, we consider the possibility of splitting a storage request onto different remote storage nodes. The latter service aims to distribute a same data sequence from one server towards multiple remote nodes assuming a limited number of network nodes have multicast capacities. These two innovative services provide a gain of 7% in terms of accepted requests when applied to the 18-node NSFnet backbone network.

Dynamic Resource Allocation in Cloud Environment Under Time-variant Job Requests

In Cloud environments, efficient resource provisioning and management present today a challenging issue because of the dynamic nature of the Cloud on one hand, and the need to satisfy heterogeneous resource requirements on the other hand. In such dynamic environments where end-users can arrive and leave the Cloud at any time, a Cloud service provider (CSP) should be able to make accurate decisions for scaling up or down its data-centers while taking into account several utility criteria, e.g., the delay of virtual resources setup, the migration of existing processes, the resource utilization, etc. In order to satisfy both parties (the CSP and the end-users), an efficient and dynamic resource allocation strategy is mandatory. In this paper, we propose an original approach for dynamic resource allocation in a Cloud environment. Our proposal considers computing job requests that are characterized by their arrival and teardown times, as well as a predictive profile of their computing requirements during their activity period. Assuming a prior knowledge of the predicted computing resources required by end-users, we propose and investigate several algorithms with different optimization criteria. However, prediction errors may occur resulting in some cases in the drop of one or several computing requests. Our proposed algorithms are compared in terms of various performance parameters including the rejection ratio, the dropping ratio, as well as the satisfaction of the endusers and the CSP.

A meta-heuristic approach for monitoring trail assignment in WDM optical networks

Optical network survivability relies on automatic fiber cut detection. Cost-effective and unambiguous fiber cut localization remains an open problem for transparent optical networks. In this paper, we focus on out-of-band monitoring techniques based on the concept of monitoring trail (m-trail) recently proposed in the literature [1]. An m-trail-based solution guarantees the exact localization of any single link failure. Two approaches have been proposed for the design of m-trails: an exact Integer Linear Program formulation and a heuristic approach based on the Random Code Assignment-Random Code Swapping (RCA-RCS) concept. Numerical applications underline two drawbacks of these approaches. The former is not scalable in terms of computational complexity. The latter provides monitoring solutions that strongly depend on the choice of the initial conditions. Thus, the final solutions can be relatively far from the optimal in the case of large networks. We propose an original meta-heuristic approach called MEMOTA that overcomes these drawbacks. Numerical results show that MEMOTA provides near optimal solutions with much shorter computing delays than the ILP formulation. In addition, unlike RCA-RCS, MEMOTA is less sensitive to initial conditions.

Impact of Traffic Predictability on WDM EXC/OXC Network Performance

Recently several investigations in the field of multilayer traffic engineering (TE) mainly focusing on random traffic demands have been carried out. In this paper, we propose different TE strategies for EXC/OXC networks including electrical traffic grooming and lightpath rerouting. Unlike existing investigations that consider only random demands for TE, we consider simultaneously scheduled and random traffic demands. Scheduled demands correspond to guaranteed services whereas random demands correspond to best-effort services. In order to outline the impact of traffic predictability on TE efficiency, we introduce the new concept of semi-random traffic demand that corresponds to random demands with predictable holding time. For a given network and a given traffic scenario, numerical simulations outline the huge gap between the rejection ratio of random demands and that of semi-random demands. In order to improve the rejection ratio of random demands, we propose to proceed for such demands to a time limited resource reservation (TLRR). The choice of the value of reservation window Delta strongly depends on the arrival and life duration processes of random demands. We show that it is possible to achieve comparable rejection ratios for both random and semi-random demands

A fast and accurate meta-heuristic for failure localization based on the monitoring trail concept

Recent advances in WDM technology enable an optical fiber to carry up to 200 wavelengths operating at 40 Gbps each. In such networks, service disruptions caused by network faults (e.g., fiber cut, amplifier dysfunction) may lead to high data losses. Therefore, it is mandatory for a network operator to be able to detect and localize such faults in the lowest possible delays. Fault detection and localization in meshed WDM networks have been deeply investigated in the literature. Numerous schemes, both exact and approximate, have been proposed to achieve fast and accurate failure localization. In this paper, we propose an original Meta-heuristic for Monitoring Trail Assignment (MeMoTA). We show through numerical results that MeMoTA provides monitoring solutions closer to optimal than existing heuristic approaches while ensuring an acceptable scalability.

Impairment-aware radio-over-fiber control plane for LTE antenna backhauling

Provisioning ubiquitous broadband wireless services necessitates the duplication of the radio equipment or/and the reduction of the size of the radio cells. However, both approaches are not economically viable for the operators. The concept of distributed antennas is proposed as an alternative solution offering low-cost broadband access. In this matter, we propose an innovative Radio Access Network (RAN) architecture called GeRoFAN for Generic Radio-over-Fiber Access Network. The GeRoFAN architecture is based on a transparent optical loop connecting low-cost 4G radio-system antennas using analog Radio-over-Fiber (RoF) technology. However, by transporting Radio Frequencies (RFs) over an Optical Channel (OC), the composite signal suffers from various physical layer impairments that may degrade the system capacity. Supported by a mathematical model reflecting the relevant optical layer limitations, we propose in this paper an innovative control plane that targets to optimize the global radio cellular capacity of GeRoFAN by efficiently mapping RFs onto OCs. This problem is formulated as an Integer Linear Program (ILP) with the objective to minimize the number of OCs subject to 4G cellular planning rules.

From Network Planning to Traffic Engineering for Optical VPN and Multi-Granular Random Demands

In this paper, we investigate the routing problem for dynamic traffic demands providing differentiated types of service across the network. Two classes of traffic are considered: QoS traffic and best-effort traffic. We propose a comprehensive offline/online routing solution that enables service providers to satisfy the demands of customer traffic flows while respecting the traffic priority. As a first step, an offline global optimization tool is used for network dimensioning. It aims to minimize the number of optical and electrical ports required in the network in order to satisfy a given set of high priority demands. Once the networks nodes are dimensioned, the network can be over-dimensioned to carry additional best-effort traffic requests. The latter set of traffic is routed by means of an online sequential optimization tool. Low priority traffic are divided into two subsets according to the requested capacity of the demands. This capacity may correspond either to an optical channel data rate (lightpath demands) or to a fraction of that rate (electrical demands). Our sequential optimization tool proceeds automatically to the grooming of the electrical demands into grooming demands. The network performance is expressed as the ratio of the number of rejected/blocked requests to the total number of requests

Impairment-aware control plane for next Generation Radio-Over-Fiber Access Networks

Radio-over-Fiber is expected to play a key role in the design of Next Generation Radio Access Networks. In this paper, we present an all-optical feeder network architecture called GeRoFAN (Generic Radio-over-Fiber Access Network) able to federate 4th generation radio mobile base stations. Thanks to its original control plane, the GeRoFAN system enables broadband ubiquitous services by dynamic management of radio resources. However, because of the transmission impairments induced by the optical layer on the transported radio channels, this broadband ubiquity may be altered. Through a study case, we highlight the necessity of an impairment-aware control plane to achieve an efficient use of the optical resources while preserving the signal quality of the transported radio channels.

OPN08-04: Traffic Routing in a Multi-Layer Optical Network Considering Rerouting and Grooming Strategies

Most of the studies dealing with dynamic lightpath establishment (DLE) consider a single type of traffic demand. These demands are in general generated randomly without any information about their life duration. In this paper, one considers for DLE a mix of preplanned traffic and of random traffic. Preplanned traffic corresponds for instance to optical virtual private networks (OVPN). Two approaches for the acceptance of random traffic demands are compared. The first approach consists in evaluating the availability of the required resources only at the instant of arrival of the random demand. The second approach consists in accepting the new random demand only if sufficient resources are available over a given period of time Delta. This time limited reservation imposes to consider at call setup preplanned resources over the period Delta. Indeed, such resources cannot be used by a random demand. The life duration of a random demand being unknown, either its extinction date occurs during the period Delta, or later. In this last case, a rerouting procedure may be necessary to enable the random demand survivability. In summary, the originality of our approach consists in providing a time-limited guaranteed service to random traffic demands instead of a pure best effort service, as it is the case in general.

Impact of Resource over-Reservation (ROR) and Dropping Policies on Cloud Resource Allocation

In Cloud environment, Cloud Providers (CP) grant access to computing and storage resources through Web-portals. Resource virtualization is a key enabler for providing such services. Thanks to virtualization, multiple Virtual Machines (VMs) can be hosted by the same Physical Machine (PM). Existing CPs, such as Amazon Web Services and Windows Azure, offer pre-configured VMs with fixed-size computing, storage and network capacities. In this context, end-users can only choose from a set of predefined VM instances offered by the CPs. However, it is expected that, in the near-future, end-users will be able to access the Cloud without any restriction on the size of the required resources. They will be charged according to the amount of resources used. In such scenario, the major problem faced by a CP is to select the appropriate PM that will host a new VM while still satisfying the end-user requirements. This resource allocation problem is similar to the well-known online Bin Packing problem. In this paper, we investigate several algorithms that were proposed to solve the Bin Packing problem, and compare them in a Cloud environment in terms of resource utilization and percentage of dropped VMs. The novel concept of Resource Over-Reservation (ROR) is introduced as a mean to reduce the percentage of dropped VMs and to improve resource utilization.