Empirical bayesian network to improve service delivery and performance dependability on a campus network

Received Sep 2, 2020 Revised Jun 1, 2021 Accepted Jun 12, 2021 An effective systemic approach to task will lead to efficient communication and resource sharing within a network. This has become imperative as it aids alternative delivery. With communication properly etched into the fabrics of today’s society via effective integration of informatics and communication technology, the constant upgrades to existing network infrastructure are only a start to meeting with the ever-increasing challenges. There are various criteria responsible for network performance, scalability, and resilience. To ensure best practices, we analyze the network and select parameters required to improve performance irrespective of bottlenecks, potentials, and expansion capabilities of the network infrastructure. Study compute feats via bayesian network design alongside upgrades implementation to result in a prototype design, capable of addressing users need(s). Thus, to ensure functionality, the experimental network uses known simulation kits such as riverbed modeler edition 17.5 and cisco packet tracer 6.0.1-to conduct standardized tests such as throughput test, application response-time test, and availability test.

. The context of complex system to be modeled The problem characteristics of the study is geared towards having a network that models as part of its components, a system that assesses availability, performance, and robustness as thus: − How do we formalize feats like availability, performance, and robustness, modelled as complexities within a campus network with respect to the network's size as a system [11]? − What combination of uncertainties feats in the parameter estimation-needs to be employed on the network to yield better quality of service, availability, and performance [11]? − We seek to factor into the network's implementation, the temporal and environmental aspects that will help yield the desired outcome [25]. − We seek to integrate qualitative and quantitative knowledge at the various levels of abstraction [26], [27]. − We seek to factor in the nature of multi-state components within the network infrastructure [28] and, − We seek to ascertain the inter-dependences between events such as failures, using simulation kits [29]. The goal (with these factors catered for) will improve service delivery via greater network resilience, improved performance, availability, and capability of the network to withstand upward scalability. Thus, will also account for system dynamics and robustness. With these feats met, we can encounter other issues such as: degradations of components, evolution of symptoms in deterioration mechanisms, impact of preventive maintenance acts on degradation, influence of environmental conditions and effects of the operation conditions on the evolution of the component states. Thus, the essence of the computation solution to be modelled in with the appropriate data to support the decision process. This implies that the uncertainty and imprecision in the system parameters not foreseen from the outset-can be accommodate with some degree of tolerance and resilience in the network.

. Principles of probability
An argument is simply a statement that has a true or false value. Reasoning is the ability to employ an argument in favor of an assumption that provided the facts, the argument can either favor the assumption that everything is either believed false or believed true. However, it is often useful to represent the facts that we believe such event is probably true/false, or it can simply be expressed that such an event is true with a probability of 0.65. Thus, the probability of an event x is true (success) or false (unsuccess) with a value in the range of 0 and 1, denoted as P(x) = 0 < x < 1 [6], [30]- [34] in (1): Basic probability rules includes: (a) all probabilities have values between 0 and 1 inclusive such that 0 ≤ P(x) ≤ 1, (b) sum of all the probabilities in sample space equals 1, (c) the probability of an event occurring equals 1, (d) the probability of the sample space is 1, (e) the probability of any event not in the sample space is 0, and lastly, (f) the probability of an event not occurring is P(x')= 1-P(x'). Probability helps us deal with problems of reasoning that involve randomness, unpredictability, and insufficient data to work out what is true/false. Thus, we have these definitions [7], [12], [30], [35], [36]: (a) probability is the likelihood that an event is successful with P, and unsuccessful with (1-P), (b) an event is the outcome of a probability experiment, (c) a sample space is the set of all outcomes in a probability event, (d) an experiment leads to well-defined results called outcomes, (e) an independent events is such that two events E1 and E2 are independent if the probability that E1 occurs, does not affect the probability of E2 occurring as well, and (f) mutually-exclusive events is such the occurrence of any events in the set E1, E2, E3…En, will automatically implies non-occurrence of the remaining n-1 events.

Bayesian networks
BN is based on bayes theorem of conditional and probabilities of random events. It is a machine learning scheme that represents data as graph mathematical structure [37]. It shows probability relations of a set of variables under uncertainty as directed acyclic graph (DAG) and conditional probability tables (CPT) of a random variable -given the occurrence of its parent nodes [38], [39]. In relation to the degree of beliefit measures the plausibility of an event given incomplete knowledge. It states that the probability of an event A conditional on another event B is given by P(A|B). It is different from probability of B conditional on Adenoted as P(B|A). It implies: (a) that Bayes Theorem is the relation between events P(A|B) and P(B|A), (b) it seeks to compute P(A|B) given data about P(B|A), and (c) its outcome uses new data to update conditional probability of event [30], [40]- [49]. So given sample space s, with mutually exclusive events (A1, A2,…,An) from s-B is an event of s with probability P(B) > 0. Thus, Bayes theorem describes the probabilities that: So that invoking the facts that: ( | ) = ( ). ( | ) -the probability then becomes (3).
BN classifiers are built using training data with structured parameter learning on probability distribution for each node on the network. It uses two learning forms: (a) structured learning (casual discovery) which learns the structure of the networks and the parameters adopted based on observed input data using either of K2, Hill climbing and Tabu-Search; and (b) probability distribution learning is achieved with algorithms like Bayes Net estimator, Bayesian model averaging (BMA) estimator and multinomial estimator. Once structure learning is complete, parameter learning completes the CPT tables for each feature in the Bayesian network [4], [50]- [52].

EXPERIMENTAL BAYESIAN NETWORK 3.1. Experimental framework/implementation
We seek to employ BN to select the appropriate parameters and compute their combination usage to improve the quality of service delivery on a campus network infrastructure, its capacity and performance. Our BN design must also address the goals, conflicts in parameter selection and estimation algorithms. We adopt the hill-climbing search on six parameters for this network on a threshold value of 0.5 [38], [39] as in Figure 2.

Feature selection, training, and rationale for choice of model
The need to optimize the number of feats employed as input parameters-stems from the fact that an increase in the number of feats used, will add to the computational complexity of the system. Thus, BN is used in selecting feats using (1-3) respectively. Input dataset [14] obtained during feasibility for improving system performance is used. We train the BN as a filter solution-since it uses known predictions and their total occurrences (as scores) to maintain a database; And, based on their occurrences, each value-combination of the parameter is considered as data assigned a criterion or probability score for its capacity of determining an improved case scenario for the network infrastructure.

Rationale for the model of choice used
The certainty-factor model was one of the most popular models of representation and manipulation of uncertain knowledge for rule-based systems. Its place has been taken by the more expressive formalisms of the bayesian belief network (BN) for the representation and manipulation of uncertain knowledge. BN is a graphic probabilistic model that represents a set of variables and their respective probabilities. It consists of: (a) a set of nodes and corresponding edges showing nodal relations, (b) edges reflects cause-effect relations, (c) the effects are not completely deterministic, and (d) the strength of an effect is modeled as a probability [6], [7], [13], [30].
BN is a directed acyclic graph whose added value is linked to probability computation of a nodal state over several variables. BN is a powerful modeling tool for complex systems as they provide modeling benefits such as: (a) BN permit users to merge and model knowledge about dynamic tasks with feedback data from experience or experts' judgment (expressed as logical rules and/or subjective probabilities), (b) it studies the behavior of a system (functional and dysfunctional analysis) and the observations therein, modelling all the complex and abstracted levels of interactions as nodal relations to help further analyze the structure, (c) it helps to establish cause-and-effect relations between nodal tasks using several sources of data to develop the model with a proviso that only a few feedback data relates to dependability [53]. Thus, we adopt an experts' judgment (network infrastructure from an international best practice view)-integrated into the model's structure. A general inference here, is to permit the propagation and diagnostic modules to incorporate the new data (evidence) gathered in a study. Thus, we use BN to allow updating of the set of events´ probabilities according to observed facts and the BN network structure. It makes the strength of this knowledge management tool [16].

RESULT FINDING AND DISCUSSION
To effectively conduct tests, scripts were designed and tested using the academic riverbed modeler edition 17.5 for the tests. The software was configured with required applications and user population-using the application cum profile configuration options, as shown in Figure 3. Figures 3(a) and 3(b) show the interfaces of the application and profile configurations respectively of the Riverbed modeler, academic edition 17.5.

Application response time test
This test is a performance metric that aims to determine the time interval between a user's request and the actual time a response is gotten, as shown in Figure 4. To achieve, the response time from a database Query, a HTTP Page, file downloads from FTP and Email Server was tracked as in Figures 4(a) and 4(b) respectively using two scenarios. In case 1, the response times for database queries was about 0.38 seconds, 0.008 seconds for email download, 0.052 seconds for file download and 0.32 seconds for hypertext transfer protocol (HTTP) page retrieval. In case 2, a longer response time was seen as it took about 0.40 seconds for database queries, 0.015 seconds for email download, 0.060 seconds for file download and 0.35 seconds for HTTP page retrieval. There was no significant difference in the response time for the various applications in both scenarios. The result concludes that the response time (even with a doubled population) is still fast and system is highly scalable as seen from Table 1.

Throughput test
Dye et al. [54] defined throughput as data transfer rate over a period in time, as shown in Figure 5. As a performance metrics, it essentially displays the effects of interference and errors on a network's capacity. Data transfer rate of the network traffic the four local area network (LAN) segments were analyzed using both scenarios as in Figure 5(a) and 5(b) respectively. In case 1, the highest data transfer rate was about 47.68 mbps and from student LAN, while the lowest is from management LAN with about 6.68 mbps. In case 2, highest throughput also from student LAN with 89.65 mbps; while the lowest again from management LAN with 11.44 mbps. This is expected. We used a multimode fiber optic cabling with a bandwidth capacity of 9.92 Gbps. As such, the effect of the highest throughput has no negative consequence on the network. With our LAN cabling capacity, optimal performance is expected.

Availability test
Ping command was used for the reachability of the different nodes. The ping sends internet control message protocol to different devices across the network. Figure 6 shows ping to different nodes with about 80% response rate (first ping), and 100% on subsequent pings shows that the different nodes were reachable.

Network security
To conduct security test for proposed network, we use the Riverbed OPNET to test the effectiveness of our firewall. The test consisted of two scenarios; the first case allowed external users to have unrestricted access to the network resources, while the second scenario restricted external users from having access the web and database servers. We used regression test to ensure that all network devices and applications from the existing network were fully incorporated and functional in the proposed network. As shown in Figure 7, existing switches were used at the access layer of the network and were found to be functional. We also tested the backup router to ascertain its capability in the event and failure of the main router. To achieve this, we have the scenario of using the existing router and switches as in Figure 7. Figure 8 shows the results from web traffic to external users and red line indicates amount of data received per second by a user, while blue line shows that no access was granted to the external users by the web server. Figure 9 shows the database server and other applications granted access to the external users with traffic still filtered. Figure 10 shows data traffic flow from the file server to the external users.

CONCLUSION
From the tests, the basic functionalities of the proposed network are working as stipulated and in tandem with the study objectives. We opine that the deployment of this network meets the needs of the stakeholders. Study employed a standardized approach to network analysis, design, optimization and testing to deliver a network that met the requirements of its stakeholders. It ensured that users' requirements were met via a top down design, which formed the basis for the network topology, applications, devices, and Int J Artif Intell ISSN: 2252-8938 Empirical bayesian network to improve service delivery and… (Arnold Adimabua Ojugo) 633 protocols chosen. Furthermore, qualitative, and quantitative methods were utilized in collection/analysis of data. The constraints of time, energy and resources that previously made it impractical to explore the benefits a network can offer any organization were considered via the implementation of four applications namely centralized data storage, file transfers, web services and e-mail services. We recommend also other possible implementation(s) such as voice over internet protocol and cloud computing for internal communication and external data storage respectively-all of which will further guarantee an optimal utilization of the network infrastructure, provide a reliable platform for data recovery and network agility required by organizations in today's competitive world.