Abstract

Today, the cloud computing paradigm enables multiple virtualized services to co-
exist on the same physical machine and share the same physical resources, hard-
ware, as well as energy consumption expenses. To allow cloud customers migrate
their services on to the cloud side, the Infrastructure Provider (InP) or cloud
data centre operator provisions to its tenants virtual networks (VNs) to host their
services. Virtual Networks can be thought of as segmenting the physical net-
work and its resources, and such VN requests (or tenants) need to be mapped
onto the substrate network and provisioned with sufficient physical resources as
per the users’ requirements. With this emerging computing paradigm, cloud cus-
tomers may demand to have highly reliable services for the hosted applications;
however, failures often happen unexpectedly in data-centers, interrupting critical
cloud services. Consequently, VN or cloud services are provisioned with redun-
dant resources to achieve the demanded level of service reliability. To maintain
a profitable operation of their network and resources, and thus achieve increased
long term revenues, cloud network operators often rely on optimizing the map-
ping of reliable cloud services. Such problem is referred to as in the literature as
“Survivable Virtual Network Embedding (SVNE) ” problem. In this thesis, the
survivable VN embedding problem is studied and a novel cost-efficient Survivable
Virtual Network Redesign algorithm is carefully designed, presented, and evalu-
ated. Subsequently, we distinguish between the communication services provided
by the cloud provider and study the problem of survivable embedding of multicast
services; we formally model the problem, and present two algorithms to reactively
maintain multicast trees in cloud data centers upon failures.