2. Marco Teórico
2.2 Competencias
A managed data center is a data center (as defined in Chapter 4) which has a management system associated with it. To form the management system, first we need to know which objects need to have a manager. From Chapter 4 we have a definition of a “data center” and the kinds of “managed objects” in the data center. Again, these “managed objects” are assumed to be the kinds of objects that can be managed within the scope of that data center model; the objects that are actually managed are specified by the administrator and are assumed to be a subset of the “managed objects” defined.
In this section we aim to define the model of our management system. Our first step is to associate a manager class with a class of managed objects.
Definition 15. For a given Data Center, DC, and MOClassesDC of DC = CMODCU
CMORacks U CMOChasses U CMOnodes U CMOSystems U CMOApps U CMOcoolers, we
define the set of Autonomic Manager Classes of DC, AMClassesDC, to be = {Ami |
Ami is a class of autonomic managers associated with CMOi where cmoi ε
managed) result in the instantiation of an autonomic manager. Pragmatically, one can think of a class of managers to be a parameterized autonomic manager that is assigned specific details about an object to manage when it is initialized. As well, the management system may or may not contain managers for all MO classes; this depends on the administrator and the architecture of the management system.
Managers in the management system can be configured to interact with one another. We could leave the administrator to explicitly define the communication among managers, but this is intractable in a data center where there may be hundreds of managers. Further,
we would like to automate this as much as possible. This is the notion of an autonomic
manager topology. For a specific set of managers their communication represents a single topology. As managers change, so does the topology. Hence, we need a general model of topologies for a data center; we do this in terms of manager classes.
A management pair in the management system has two manager classes and priority. The pair of classes means that managers in those classes can communicate with each other. The priority determines which Autonomic Manager Class (AM class) in the pair is a privileged, if there is one. AMs of the privileged class can be thought of as “managing” AMs of the other AM class. If there is no privileged AM class, then AMs in the two classes are essentially peers. A privileged AM can, in essence, take actions to influence the behavior of an AM in the other class, including, for example, changing the policies of the subordinate AM. These actions are done through communications between the managers; we will elaborate on this later.
Definition 16. For a given Data Center, DC, the Set of Managed Object Classes, MOClassesDC,, and a Set of Autonomic Manager Classes, AMClassesDC, a
Management Pair set, AMP, is defined as:
For example, if (AM1, AM2, priority) is a management pair, then if the priority=1 that
means that AMs of AM1class are privileged can change the settings or behavior of AMs
of AM2.
A management topology is then defined as a set of manager classes and the management pair set corresponds to vertexes and edges in a graph.
Definition 17. For a given Data Center, DC, Set of Managed Object Classes MOClassesDC and a Set of Autonomic Manager Classes, AMClassesDC, a
Management Topology TDC for DC is a graph G of <V, E>
Where:
V= {vi| vi
∈
AMClassesDC }. E = {e| e = {(Ai, Aj, priority)| Ai and Aj
∈
AMClassesDC}}A management topology actually defines the relationship between different management classes attached to managed object classes. In specifying the topology, one specifies constraints on the relationships that can occur between different managers. In practice, this topology will be instantiated during data center administration and operations resulting in a graph connecting managers that is consistent with this topology.
Having defined a management topology, data center, classes of managed objects, and classes of managers, we can define a managed data center, as follows:
Definition 18. A managed data center MDC is defined as <DC, T>
Where:
DC = <Systems, Racks, ThermalModel>, is a data center with Set of Managed Object Classes, MOClassesDC, Set of Autonomic Manager Classes,
The definition of a Managed Data Center defines the management framework. In practice, depending on the administrator, within that framework not all managed objects need to be considered. That is, one can think of the framework as defining what can be
managed and relationships among management elements, while what is actually managed
can be dictated by the administrator. Abstraction level in the management system is illustrated in Figure 5-1. At the first level, we have set of possible manageable objects; here we assume that these are objects which are to be managed as well as autonomic managers that have been or could be defined. We assume that these are grouped into classes (the second level). This allows objects with similar management properties, e.g. sensors, actuators, to have a single definition and that properties of a specific object can be instantiated. Third level is the autonomic management class level; each AM class corresponds to one or more MO classes. This allows an autonomic manager to be instantiated, i.e., its properties specified, when an object is instantiated. Allowable connections among autonomic managers are defined as autonomic manager pairs at the next level. The connections between classes of autonomic managers define the autonomic manager topology of the management system.