Information Technology Reference
In-Depth Information
2
Background
ff
Clouds [14] are the current emerging trend in delivering IT services, and o
er to end-
users a variety of services covering the entire computing stack. Scientists in general and
PSEs users in particular can completely customize their execution environment, thus
deploying the most appropriate setup for their experiments. Another related important
feature is the ability to scale up and down the computing infrastructure according to
PSEs resource requirements.
In the next subsections we describe the federated Cloud basics (Subsection 2.1),
and introduce the classical SI-based algorithms (Subsection 2.2), the core optimization
techniques of the schedulers implemented in this work at the infrastructure level.
2.1
Federated Clouds
Federated Clouds [15] consist of infrastructures with physical resources belonging to
di
erent architectures and
levels of coupling among federated datacenters. Federated Clouds also make use of
brokers
to meet the needs of their participating organizations. A broker is an entity
which keeps a queue of requests from a particular user that need to be provisioned by
a datacenter. In the context of this work, where a user runs PSEs, only one broker is
associated with that user.
Clouds allow the dynamic scaling of users applications by the provisioning of com-
puting resources via
machine images
, or VMs. In order to achieve good performance,
VMs have to fully utilize its services and resources by adapting to the Cloud dynam-
ically. Proper allocation of resources must be guaranteed so as to improve resource
usefulness [15].
For running applications in a Cloud, resources are scheduled at three levels (Fig-
ure 1): Broker level, Infrastructure level, and VM level. At the broker level, di
ff
erent Cloud providers. A federated Cloud could involve di
ff
ff
erent
policies can be implemented in order to serve users. Some examples are policies consid-
ering the influence of network interconnections among Cloud datacenters or monetary
cost of hosts that compose them [1]. Furthermore, the scheduler at this level can decide
to deploy the VMs in a remote Cloud when there are insu
cient physical resources in
the datacenter where the VM creation was issued. Secondly, once a datacenter
provider
has been selected by a broker, at the infrastructure level, the VMs are allocated into real
hardware through a VM scheduler. Finally, at the VM level, by using job scheduling
techniques, jobs are assigned for execution into virtual resources (the allocated VMs).
Figure 1 illustrates a Cloud where one or more users are connected via a network and
require the creation of a number of VMs for executing their experiments, i.e., a set of
jobs. As can be seen in the Figure 1, a broker is created for each user that connects to the
Cloud. Each broker knows who are the providers that are part of the federation through
network interconnections -the relation of each broker is colored with green and blue
dotted lines-. In addition, the Figure 1 illustrates how jobs sent by
User N
are executed
in the datacenter of
Cloud Provider 2
. At the right of this provider -inside the dotted
Cloud- the intra-datacenter scheduling activities are depicted, i.e., at the infrastructure
level and the VM level.
/
