After correlating commands to the related interfaces we extract sequential pat

After correlating commands to the related interfaces

This preview shows page 27 - 29 out of 169 pages.

customer-facing interfaces that connect to different sites of the same enterprise. After correlating commands to the related interfaces, we extract sequential pat- terns, representing combinations of commands that are executed together to fulfill different management tasks. From the data we also observe that in addition to com- mands that lead to persistent configuration changes, virtually all management tasks also involve status-checking commands that do not change the configuration, but allow the operator to verify network states, which largely determines the follow-up actions. Such operations are critical in operational procedures, but largely ignored by existing studies. Finally, we modelnetworkmanagement operationsusingautomaticallygenerated deterministicfiniteautomata(DFA),whereastaterepresentstheconfiguredbehavior ofaninterface andanedgeindicates theoperationsperformedontheinterface either to fulfill a specific task or to check the status of the network. The DFA model not only allows us to capture common management tasks, but also gives us the ordering and dependency information among those tasks. Containing the information about thetemporalprogressionofnetworkmanagementunderdifferentnetworkconditions, the DFA model provides a dynamic view of how large networks are managed today. Basedonthisunderstanding,bettertoolsforautomatingnetworkmanagementcanbe 14
built. We argue that composing DFAs is a better network management abstraction, which enables operators to reason about the operational state of the network. This rest of this chapter is organized as follows. Section 2.2 describes the data sourcesandtheinitialprocessingstepstakentofacilitatelateranalysis. InSection2.3, wepresentthemainanalysisandresults. Thekeymethodistoexploitthereferential relationships to associate commands related to individual interfaces. Section 2.4 describes howthe DFAmodel isused tocapture network dynamics and the potential usages of such a model. Finally, section 2.5 summarizes this chapter. 2.1 Motivation for Using a DFA Model DFA stands for deterministic finite automaton [71], which consists of a finite set of states and transitions that depend on input symbols at each step. A number of factors imply that DFA is a good abstraction to model network operations: 1. A network usually transits from one state to another after a series of manage- ment operations are performed on it. 2. The state of the network limits the possible operations to be performed. 3. There are a limited number of network states and operations. (1) is clearly true when the operations make configuration changes, so that the resulting network functionality differs from before. Even when the operations are for checking network status only, (1) is also true from the operator or management systems’ perspective. Usually, astatus-checking operationwould reveal certain prop- erties of the network, e.g., testing if a customer interface is properly configured, or if a backbone interface still has traffic flowing through. The result of these checking operations would change the view of the network. (2) is also true because there is usually ordering and dependency among the operations. For example, an operator

  • Left Quote Icon

    Student Picture

  • Left Quote Icon

    Student Picture

  • Left Quote Icon

    Student Picture