Information Technology Reference
In-Depth Information
the answer, i.e., the period of time in which a user requests a number of VMs to execute
its PSE, and the time in which all the entire PSE-jobs finish their execution. Concep-
tually, the scheduling problem to tackle down can be formulated as follows. A PSE is
formally defined as a set of
N
n
independent jobs, where each job corresponds
to a particular value for a variable of the model being studied by the PSE. The jobs are
distributed and executed on the
v
VMs issued by the user. With the goal of minimizing
the response time, the need to implement strategies to select the appropriate datacenters
in which to place the VMs arises. For example, the most suitable datacenter might be
the one that provides the lowest communication latency to a broker when this latter asks
about the availability of physical resources. Latency is due to delays by packets mov-
ing over the various networks between the end user computer and the geographically
distributed Cloud datacenters. One way to mitigate the e
=
1
,
2
,...,
ff
ects of such latencies is to
choose a datacenter which operates with a fast and e
cient internal network and plenty
of capacity.
The proposed scheduler proceeds as follows. Firstly, at the broker level, a datacenter
is selected by a policy that takes into account network interconnections and
or network
latencies. Secondly, at the infrastructure level, by means of a VM scheduler, user VMs
are allocated in the physical resources (i.e., hosts) belonging to the selected datacenter
at the broker level. When there are no available hosts in the datacenter to allocate the
VMs, a new datacenter is selected at the broker level. Finally, at the VM level, a policy
for assigning user jobs to allocated VMs is also used (currently we use FIFO).
/
4.1
Scheduler at the Broker Level
The scheduler at the broker level is executed both to select the first datacenter to allo-
cate the VMs, which are managed by the scheduler implemented at the infrastructure
level, as well as each time such datacenter is not able to perform the allocation of VMs
anymore. At present, the policies implemented at this level are:
-
Lowest-Latency-Time-First (LLTF), maintains a list of all network interconnected
datacenters sorted by their latencies. Each time a user requires a number the VMs
to execute their PSE, this policy is responsible for selecting first the datacenter with
the lowest latency in the list. Then, whenever a datacenter has no more physical
resources to allocate VMs, then the algorithm selects the next datacenter in the list.
-
First-Latency-Time-First (FLTF) selects the first datacenter from a list sorted ran-
domly, containing all network interconnected datacenters to which a user can access
and allocate his VMs. When the selected datacenter has no more available physical
resources to allocate VMs, the algorithm selects the next datacenter in the list.
-
Latency-Time-In-Round (LTIR) maintains a list of all network interconnected dat-
acenters that make up the Cloud, sorted by increasing latency, and assigns each VM
required by the user to a datacenter from the list in a circular order.
4.2
Scheduler at the Infrastructure Level
To implement the Infrastructure level policy, the SI algorithms proposed in [18] are
used. Below we describe these algorithms.
