The increasing demand for cloud computing drives the expansion in scale of datacenters and their internal optical network, in a strive for increasing bandwidth, high reliability, and lower latency. Optical transceivers are essential elements of optical networks, whose reliability has not been well-studied compared to other hardware components. In this paper, we leverage high quantities of monitoring data from optical transceivers and OS-level metrics to provide statistical insights about the occurrence of optical transceiver failures. We estimate transceiver failure rates and normal operating ranges for monitored attributes, correlate early-observable patterns to known failure symptoms, and finally develop failure prediction models based on our analyses. Our results enable network administrators to deploy early-warning systems and enact predictive maintenance strategies, such as replacement or traffic re-routing, reducing the number of incidents and their associated costs.

In large-scale datacenters, memory failure is one of the leading causes of server crashes, and uncorrectable error (UCE) is the major fault type indicating defects of memory modules. Existing approaches tend to predict UCEs using Correctable Errors (CE). However, bit-level CE information has not been completely discussed in previous works and CEs with error bit patterns are strongly correlated with UCE occurrences. In this paper, we present a novel Hierarchical Intelligent Memory Failure Prediction (HiMFP) framework which can predict UCEs on multiple levels of the memory system and associate with memory recovery techniques. Particularly, we leverage CE addresses on multiple levels of memory, especially bit-level, and construct machine learning models based on spatial and temporal CE information. Results of algorithm evaluation using real-world datasets indicate that HiMFP significantly enhances the prediction performance compared with the baseline algorithm. Overall, Virtual Machines (VM) interruptions caused by UCEs can be reduced by around 45% using HiMFP.

Accurate, timely Root Cause Analysis (RCA) is essential to successful IT operations as a primary step to incident remediation. RCA automation using data mining techniques in large heterogeneous systems is, however, a challenging task, because it requires correlating multimodal information across various data sources. An increasing number of services are migrating to structured logging to enable automated monitoring and debugging of complex large-scale systems. In this paper, we leverage structured logs and association rule mining (ARM) to automate RCA. We propose the LogRule algorithm, which automatically analyzes structured logs to generate a list of explanations for an event of interest. It achieves 0.921 F1-score for the diagnosis task, while computing results 37x faster compared to the state-of-the-art solution based on FP-growth, making it a time-efficient, accurate, and interpretable ARM-based RCA algorithm. Evaluation results show that LogRule enables RCA in complex multidimensional datasets, where the execution time of the current state-of-the-art algorithm is prohibitively large.

Advances in software defined and data center networks have enabled network virtualization. Virtual network embedding increases resources utilization and reduces cost of network deployment. Its performance depends on embedding algorithms and data center network topologies. In this paper, we evaluate performance of virtual network embedding algorithms based on acceptance ratio, revenue to cost ratio, and node and link utilizations by simulating virtual network embeddings on Spine-Leaf, Three-Tier, and Collapsed Core data center network topologies.

Network virtualization helps overcome shortcomings of the current Internet architecture. The virtualized network architecture enables coexistence of multiple virtual networks (VNs) on an existing physical infrastructure. VN embedding (VNE) problem, which deals with the embedding of VN components onto a physical network, is known to be -hard. In this paper, we propose two VNE algorithms: MaVEn-M and MaVEn-S. MaVEn-M employs the multicommodity flow algorithm for virtual link mapping while MaVEn-S uses the shortest-path algorithm. They formalize the virtual node mapping problem by using the Markov decision process (MDP) framework and devise action policies (node mappings) for the proposed MDP using the Monte Carlo tree search algorithm. Service providers may adjust the execution time of the MaVEn algorithms based on the traffic load of VN requests. The objective of the algorithms is to maximize the profit of infrastructure providers. We develop a discrete event VNE simulator to implement and evaluate performance of MaVEn-M, MaVEn-S, and several recently proposed VNE algorithms. We introduce profitability as a new performance metric that captures both acceptance and revenue to cost ratios. Simulation results show that the proposed algorithms find more profitable solutions than the existing algorithms. Given additional computation time, they further improve embedding solutions.

Data centers are core infrastructure of cloud computing. Network virtualization in these centers is a promising solution that enables coexistence of multiple virtual networks on a shared infrastructure. It offers flexible management, lower implementation cost, higher network scalability, increased resource utilization, and improved energy efficiency. In this paper, we consider switch-centric data center network topologies and evaluate their use for network virtualization by comparing Deterministic (D-ViNE) and Randomized (R-ViNE) Virtual Network Embedding, Global Resource Capacity (GRC), and Global Resource Capacity-Multicommodity (GRC-M) Flow algorithms.

Network visualization enables support and deployment of new services and applications that the current Internet architecture is unable to support. Virtual Network Embedding (VNE) problem that addresses efficient mapping of virtual network elements onto a physical infrastructure (substrate network) is one of the main challenges in network virtualization. The Global Resource Capacity (GRC) is a VNE algorithm that utilizes for virtual link mapping a modified version of Dijkstra's shortest path algorithm. In this paper, we propose the GRC-M algorithm that utilizes the Multicommodity Flow (MCF) algorithm. MCF enables path splitting and yields to higher substrate resource utilizations. Simulation results show that MCF significantly enhances performance of the GRC algorithm.

Network visualization enables coexistence of multiple virtual networks on a shared infrastructure without requiring unified protocols, applications, and control and management planes. Recent approaches such as Software Defined Networking have enabled cloud service providers to offer virtualized network services that require embedding virtual network requests in data centers. In this paper, we employ R-Vine, D-Vine, and Global Resource Capacity (GRC) algorithms to perform a series of virtual net work embeddings on BCube and Fat-Tree substrate networks. We compare these two data center network topologies to determine the topology that is better suited for virtual network embeddings. Simulation results show that the Fat-Tree network is capable of hosting additional virtual network requests, resulting in higher substrate node and link utilization.

The exponential growth of the Routing Information Base (RIB) of the Internet's Default-Free Zone (DFZ) routers has raised concerns about non-scalability of the current Internet's routing architecture. The main reason is that Internet addresses currently carry information about both the identity and location (physical connection point) of devices connected to the Internet. The Locator/ID Separation Protocol (LISP) has been introduced to potentially remedy this non-scalability by splitting the location and identity of devices. In this paper, we present the architecture of a deployed testbed that is multihomed using LISP. We investigate LISP performance as a multihoming solution in terms of load balancing and traffic routing in the case of link failures.

Deflection routing is employed to ameliorate packet loss caused by contention in buffer-less architectures such as optical burst-switched networks. The main goal of deflection routing is to successfully deflect a packet based only on a limited knowledge that network nodes possess about their environment. In this paper, we present a framework that introduces intelligence to deflection routing (iDef). iDef decouples the design of the signaling infrastructure from the underlying learning algorithm. It consists of a signaling and a decision-making module. Signaling module implements a feedback management protocol while the decision-making module implements a reinforcement learning algorithm. We also propose several learning-based deflection routing protocols, implement them in iDef using the ns-3 network simulator, and compare their performance.

Contention is the main source of information loss in buffer-less network architectures where deflection routing is a viable contention resolution scheme. In recent years, various reinforcement learning-based deflection routing algorithms have been proposed. However, performance of these algorithms has not been evaluated in larger networks that resemble the autonomous system-level topology of the Internet. In this paper, we compare performance of three reinforcement learning-based deflection routing algorithms by using topologies generated with Waxman and Barabási-Albert algorithms. We examine the scalability of deflection routing algorithms by increasing the network size while keeping the network load constant.

Border Gateway Protocol (BGP) is the core component of the Internet's routing infrastructure. Abnormal routing behavior impairs global Internet connectivity and stability. Hence, designing and implementing anomaly detection algorithms is important for improving performance of routing protocols. While various machine learning techniques may be employed to detect BGP anomalies, their performance strongly depends on the employed learning algorithms. These techniques have multiple variants that often work well for detecting a particular anomaly. In this paper, we use the decision tree and fuzzy rough set methods for feature selection. Decision tree and extreme learning machine classification techniques are then used to maximize the accuracy of detecting BGP anomalies. The proposed techniques are tested using Internet traffic traces.

In this paper, we introduce a predictive Q-learning deflection routing (PQDR) algorithm for buffer-less networks. Q-learning, one of the reinforcement learning (RL) algorithms, has been considered for routing in computer networks. The RL-based algorithms have not been widely deployed in computer networks where their inherent random nature is undesired. However, their randomness is sought-after in certain cases such as deflection routing, which may be employed to ameliorate packet loss caused by contention in buffer-less networks. We compare the proposed algorithm with two existing reinforcement learning-based deflection routing algorithms. Simulation results show that the proposed algorithm decreases the burst loss probability in the case of heavy traffic load while it requires fewer deflections. The PQDR algorithm is implemented using the ns-3 network simulator.

The Border Gateway Protocol (BGP) is the de facto Internet routing protocol. Various aspects of the BGP protocol have been analyzed using mathematical and experimental approaches. Formal verification of BGP specification validates whether or not a specific set of requirements is satisfied. In resent years, the probabilistic behavior of BGP has been explored. The size of routing tables has been modeled as a stochastic process that changes over time according to some probability distribution function. Hence, the verification of BGP may also be probabilistic in nature due to its randomized behavior. In this paper, we present a probabilistic model checking approach to analyze BGP convergence properties that may be employed to automate the BGP convergence analysis.