A protocol-independent container network observability analysis system based on eBPF

This thesis provides a detailed analysis of the effects of TCP and UDP traffic over a LAN and WAN medium. In addition, it also analyses some real time applications like audio, video and web browsing that is affected by TCP traffic while sharing a bottleneck node and/or link resources. As network industry is growing continuously, the network administrator should be aware of TCP and UDP traffic that is traversing through their network. The analysis and monitoring of the traffic is crucial as it directly affects the performance of the network. Finding a cause for the poor performance of the network is quite important because it gives an idea to troubleshoot and resolve the issues effectively. In this thesis, we have created topologies using Cisco’s Virtual Internet Routing Lab (VIRL) [7]. They replicate an organizational infrastructure with client-server environment in LAN and WAN. Routing protocols such as OSPF and BGP are used to mimic the real world internet. A built-in LXC-iperf tool is used to generate the TCP and UDP traffic. During the generation of the traffic, various parameters are changed or controlled to see their effect on the network performance. As a learning and informative research, this thesis considers several Quality of Service (QoS) parameters that characterize the performance of an overall network. In particular, this thesis obtains packet loss, throughput, and jitter as QoS parameters when the resource has both TCP and UDP traffics simultaneously. We have examined, the effects on (i) TCP throughput and packet loss and (ii) UPD packet loss and jitter of (a) real time audio, (b) video, and (c) web browsing applications. These parameters examine how the traffic be manipulated to keep minimum packet loss, minimum jitter and maximum throughput. It is needless to say that all are competing with each other (TCP/UDP traffic) for sharing bottleneck resources. We have used a sample time of 15 seconds for each of our experimental results presented in this thesis. Our analysis shows that the best performance of the real time video and audio application is obtained when we select large size packet but its size being less than MTU of the link (without reducing the data rates). Similarly, in case of web browsing, we notice that throughput increases by increasing the window size and decreasing the latency. Efficient outcomes with the traffic analysis are achieved only if the experiments are carried out with adequate amount of attention.

Stochastic Geometry has attained massive growth in modelling and analysing of wireless network. This suits well for analysing the performance of large scale wireless network with random topologies. Analytical framework is established to evaluate the performance of the network. Here we have created a mathematical model for uplink analysis and the gain of uplink and downlink is obtained. Then ad-hoc network architecture is designed and the performance of the network is compared with the traditional method. Finally, a new scheduling algorithm is developed for cellular network and the gain parameter is quantified with the help of Stochastic Geometry tool. The accuracy is acquired from extensive Monte Carlo simulator.

A system designed for supporting the network performance analysis and forecast effort is presented, based on the combination of offline network analysis and online real-time performance forecast. The off-line analysis will perform analysis of specific network node performance, correlation analysis of relative network nodes performance and evolutionary mathematical modeling of long-term network performance measurements. The online real-time network performance forecast will be based on one so-called hybrid prediction modeling approach for short-term network, performance prediction and trend analysis. Based on the module design, the system proposed has good intelligence, scalability and self-adaptability, which will offer highly effective network performance analysis and forecast tools for network managers, and is one ideal support platform for network performance analysis and forecast effort.

Analysis of network and application performance is traditionally based on mathematical models derived from first principles. Such models are usually complex, difficult to use, and do not exhibit satisfactory accuracy. This paper describes a new way of achieving high accuracy in predicting network performance. The new solution is built on a unique combination of network-component models and probabilistic analysis. The resulting model-based approach provides high fidelity for decision making in both monitoring and prediction modes. This paper discusses key features of the technique and illustrative results. Resulting capabilities are implemented in a system called NetPredictor/sup TM/, which can monitor and predict quality of service (QoS) and compliance with service-level agreements (SLAs). It also empowers users to perform service-level management (SLM) and to make informed decisions about effects of contemplated changes in a network. Networks can be combinations of LANs, WANs, intranets, and the Internet.

The router, as a gateway in the network, plays a vital role. If its functionality is disrupted by Denial of Service (DoS) attacks, which pose a serious cyber threat and have destructive effects by overwhelming the target with large amounts of traffic, it directly impacts network performance. This research aims to analyze DoS attacks on network devices based on MikroTik routers and switches, which become targets that disrupt network traffic. These attacks lead to a significant increase in data traffic, affecting the stability and performance of the network. Through analysis using network forensic methods, this study evaluates the performance of MikroTik firewalls in mitigating DoS attacks by assessing the traffic received and rejected based on applied rules. The results show that the firewall can handle most attacks; however, some suspicious packets still bypass it. This study emphasizes the importance of enhancing security systems and mitigating DoS attacks to maintain optimal network performance in the future.

Analytical modeling techniques can be used to study the performance of optical multistage interconnection network (OMIN) effectively. MINs have assumed importance in recent times, because of their cost-effectiveness. An N/spl times/N MIN consists of a mapping from N processors to N memories, with log/sub 2/N stages of 2/spl times/2 switches with N/2 switches per stage. The interest is on the study of the performance of unbuffered optical multistage interconnection network using the Banyan network. The uniform reference model approach is assumed for the purpose of analysis. In this paper we apply the analytical modeling approach on an N/spl times/N OMIN with limited crosstalk (conflicts between messages) unto (log/sub 2/N-1). Messages with switch conflicts satisfying the constraint of (log/sub 2/N-1) are allowed to pass in the same group, but in case of a link conflict, than that message is routed in a different group. The analysis is performed by calculating the bandwidth of the network operating under a load l and allowing random traffic and using a greedy routing strategy. A number of equations are derived using the theory of probability and the performance curves are plotted. The results obtained show that the performance of the network improves by allowing limited crosstalk in the network.

The design of Medium Voltage (MV) Overhead Lines (OHL) is often radial in nature which makes back-feeding difficult and these networks are often long which increases their exposure to faults. This has resulted in poor network reliability. Customers are mainly affected by faults on the MV network, to which particular attention has to be paid. Permanent faults (involving equipment damage) have negative impact on customers since they experience outages or interruptions while the equipment is repaired or replaced. The impact on customers increases when these outages are long. Customers continue to demand higher levels of uninterrupted supply from the distribution network; therefore it is necessary to implement self-healing Smart Grid technologies. One such technique is Fault Location, Isolation and Service Restoration (FLISR). The main drive towards implementing FLISR is to improve Eskom's network performance, increase electricity sales, improve flexibility and reduce the impact on the economy of outages. This paper discusses various ways to implement FLISR and will also focus on case studies on real distribution networks and look at the issues associated with closing a remotely controlled or automatically operated Normally Open (N/O) point for back-feeding. This paper aims to show how automated service restoration can reduce switching and restoration time from hours to minutes.

In order to satisfy the requirement of extremely high data rate in traffic hotspot regions, millimeter wave (mmWave) has attracted significant attention in wireless communication networks. While the coverage performance of mmWave networks based on the distribution of signal-to-interference-plus-noise ratio (SINR) has been widely studied, it provides only very limited information on the link reliability. In this paper, we provide a fine-grained performance analysis of the mmWave networks with hotspots. Specifically, we first establish a general and tractable framework to investigate the performance of mmWave networks using the Poisson cluster process integrated with several features of the mmWave band. Both open and closed association strategies are considered. To show what fraction of users in the networks achieves target reliability when the SINR is given, we derive the tier association probability and the moments of the conditional SINR distribution, based on which the exact meta distributions of SINR are given. Interestingly, in clustered mmWave networks, the widely used standard and generalized beta approximations do not work well when the blockage effect is severe. To resolve this issue, we provide a modified approximation by scaling the standard beta distribution, which is shown to be closer to the exact results. We conduct extensive simulations to study the impact of mmWave and deployment features on the performance of clustered mmWave networks. Numerical results reveal that the optimal scattering variance of mmWave base stations scales with the cluster size to maximize the number of concurrent reliable links, and increasing the antenna directivity results in more reliable communications than elevating the transmit power of mmWave base stations.

In the face of massive application containerization scenarios, traditional network performance anomaly detection has been unable to meet the requirements of cloud native scenarios in terms of granularity, performance, and security. Traditional network performance anomaly detection can only monitor the data exposed by the kernel. Coupled with system virtualization, it is difficult to monitor container-granular performance data, and it is easy to threaten system security by modifying the system and applications extract data. In response to these problems, this work proposes a non-intrusive container network performance anomaly detection model. The model is based on extended Berkeley Packet Filter (eBPF) technology, through non-intrusive collection of process granularity data in the Linux kernel, to obtain container granularity network performance data, combined with machine learning classification methods, to identify whether the container network performance is abnormal. After testing, the model can accurately extract container network performance data, the overall system overhead is small, and the accuracy of identifying abnormal container network performance reached 98.1%.KeywordsNetwork performanceAnomaly detectioneBPFMachine learning

The Global Positioning System (GPS) applications are highly vulnerable to the ionospheric space weather effects. Modelling and forecasting the ionospheric effects such as time delays for GPS signals are important for real time alerts of space weather effects on GPS services. In the present work, the performance of Neural Networks (NN) model is compared with International Reference Ionosphere (IRI) models. The ionospheric Total Electron Content (TEC) observations have been collected during 2015 year, descending phase of 24th solar cycle, over three Indian low latitude GPS stations namely, Bengaluru (Geographic coordinates: 13.02°N and 77.57°E), near to geomagnetic equator, Guntur (Geographic coordinates: 16.37°N and 80.37°E), which is at Equatorial Ionization Anomaly (EIA) and Lucknow (Geographic coordinates: 26.83°N and 80.92°E), which is beyond EIA region. The performance of NN model in predicting the ionospheric TEC values is compared with IRI (IRI-2012 and IRI-2016) models during test period, October–December 2015 over three Indian low latitude regions. It is observed that IRI models (IRI-2012 and IRI-2016) have shown more the temporal differences with GPS-VTEC during sunrise hours compared to sunset hours over three low latitude regions. The performance of IRI-2016 model has apparently better than IRI-2012 model. However, it is observed that IRI-2016 model has large discrepancies over Bengaluru and Guntur station due to high VTEC fluctuations at equatorial and low latitudes. The NN models are well predicted the measured diurnal mean VTEC variations with the less errors, ±5 TECU but the differences of IRI models are ±15 TECU over all the three stations. Later, GPS-data for 10 years, 2009–2018, is collected over Bengaluru station during 24th solar cycle. The performance of NN model is validated during 2016, 2017 and 2018 years over Bengaluru GPS station. The error measurements and experimental results reported that the measured GPS-VTEC values are well predicted by NN model compared to IRI-2016 model over equatorial and low latitude GPS stations.

Regularly increasing number of the subscribers in a specific area has compelled the network providers to search for different ideas to vigorously optimise the sector for increase of coverage & capacity with the quality of service in consideration. The researches on smart antenna systems have shown that they are sufficient for providing vigorous network performance. Network providers are required to calculate the gain representing the performance that can be find out by using this smart antenna system and a detailed estimation of its propagation results is a most priority factor of consideration for the dynamic improvement of network performance. Here, in this paper we represents a detailed performance analysis of wireless network performance with the use of this smart antenna system.

In this paper, we discuss the phase change flow of network and the change law of the network performance under different traffic input rates. Based on the stability conditions of the and the system, the approximate expressions of the phase change flow of the network traffic are We also derive the analytical results of the queue length, average delay, network throughput and performance indicators and the deviation degree of the network behavior on the stability of the network behavior. At the same time, we carry out the simulation of the data according to the actual situation, and the simulation results show that (1) The global behavior of heterogeneous network heterogeneous topology has local characteristics from the free flow regime to the congestion state. It is a feasible method to improve the network performance by transforming the network topology improve the value of the network traffic flow. (3) As the network load increases, the global behavior of the network is directly from the free flow state to the global congestion state. There is no state between the two states. (4) When the flow rate exceeds the global phase transition, the queue length and the average delay of the packet are increased dramatically, and the maximum value of network throughput is quickly reduced to zero.

Container-based virtualization offers advantages such as high performance, resource efficiency, and agile environment. These advantages make Internet of Things (IoT) device management easy. Although container-based virtualization has already been introduced to IoT devices, the different network modes of containers and their performance issues have not been addressed. Since the network performance is an important factor in IoT, the analysis of the container network performance is essential. In this study, we analyze the network performance of containers on an IoT device, Raspberry Pi 3. The results show that the network performance of containers is lower than that of the native Linux, with an average performance difference of 6% and 18% for TCP and UDP, respectively. In addition, the network performance of containers varies depending on the network mode. When a single container runs, bridge mode achieves higher performance than host mode by 25% while host mode shows better performance than bridge mode by 45% in the multicontainer environment.

To deal with challenges of enhanced user density, high data rate demand and coverage, conventional macro cellular network architecture is evolving to irregular and complex Heterogeneous Networks (HetNets) through dense deployment of low power Base Stations (BSs) such as pico and femto BSs. Various parameters and their effects are considered while analyzing the performance of HetNets. In this paper we investigate the impact of macro and pico BSs separation on the network performance. We consider picocell within the association region of a macrocell and investigate the association region of picocell, average rate, and Energy Efficiency (EE) of the network. The results show that network performance is improved with increase in macro and pico BSs separation within the association region of a macrocell.

Nodes in Mobile Ad-hoc Networks (MANETs) are battery powered and hence have limited life time. Due to excessive utilization a node may die which can result in energy depletion problem and thereby affect overall network performance. Local rate of energy consumption based on application (e.g.: audio, video, word, etc) can be monitored. We can estimate the remaining battery life (residual energy in a node) based on the application (rate of consumption). Thus early detection and avoidance of energy depletion problem is possible based on monitoring of power (rate of energy consumption) and remaining battery life (residual energy in a node). To avoid energy depletion problem due to excessive energy consumption at a higher rate, we introduce the mobile relays. These mobile relays reduce the power consumption in the node by balancing the load. As and when the rate of energy consumption based on the application returns to efficient levels, the mobile relay is released so that it can be made available for other critical nodes. This technique can be applied to any protocol used in Mobile Ad-hoc Networks (DSDV, DSR, AODV, etc). A history of the use of the Mobile Relay is maintained for further analysis of network performance.