Search

Adopting reinforcement learning in the network scheduling area is getting more attention than ever because of its flexibility in adapting to the dynamic changes of network traffic and network status. In this study, a timeslot scheduling algorithm for traffic, with similar requirements but different priorities, is designed using a double deep q-network (DDQN), a reinforcement learning algorithm. To evaluate the behavior of the DDQN agent, a reward function is defined based on the difference between the estimated delay and the deadline of packets transmitted at the timeslot, and on the priority of packets. The simulation showed that the designed scheduling algorithm performs better than existing algorithms, such as the strict priority (SP) or weighted round robin (WRR) scheduler, in the sense that more packets arrived within the deadline. By using the proposed DDQN-based scheduler, it is expected that autonomous network scheduling can be realized in upcoming frameworks, such as time-sensitive or deterministic networking.

A packet transport network (PTN) for optical carrier Ethernet needs to be reliable even in a case that a fault occurs on a data path. To survive from the network fault, protection switching which sends data traffic to a pre-established backup path can be implemented on optical PTN systems. Linear protection switching in a multi-protocol label switching - transport profile (MPLS-TP) environment requires to be performed with sub-50 ms for multiple faults on data paths (i.e. tunnels). The linear protection switching for the recovery from multiple faults at almost ten tunnels can be processed on CPU. In this study, a packet transport layer protection switching integrated circuit (PPSI) is developed to increase the number of multiple protection switchings (e.g. 4000 + tunnels) with satisfying the strict requirement on sub-50 ms switching time, by processing on a field-programmable gate array (FPGA). The performance of this FPGA implementation is compared with that of the processed on CPU. Furthermore, this study investigates four message forwarding methods of set-bridge-and-selector and automatic protection coordination protocol messages to enhance the performance of the PPSI. The four methods are evaluated using an FPGA test-bed, in terms of switching time and the number of tunnels.

Shared mesh protection (SMP) is a protection mechanism that recovers traffic delivery against failures on a working path as rapidly as linear protection. In addition, it allows for resource sharing among the protection paths with different endpoints in a mesh network by coordinating the use of shared resources when multiple protection paths compete for these resources. Owing to limited protection resources and the preemption in consequence of priority comparison, the SMP introduces a new optimization problem whose objective is to find the optimal priority assignment that maximizes the number of protected services against any failure case for a given SMP network. In this paper, we recognize that the problem of finding the optimal solution is NP-hard and propose min-degree and capacity-density approximation algorithms, whose approximation ratios are presented with the maximum degree of services and the maximum capacity of segments of an SMP network. Moreover, we propose a depth-based algorithm that always finds the optimal solution for a hypertree-shaped SMP network. We compare the performance of the proposed algorithms in terms of the protection ratio, which we define in this paper.

Summary Shared mesh protection (SMP) protects service traffic on a working path and requires coordinated use of the shared resources on a protection path when the traffic is routed through the prot...

A carrier suffering a communication disaster may be unable to maintain network services using its own protection resources. In this case, it could avoid service disruptions by using other carriers' resources. However, along with national regulatory efforts requiring to coordinate multiple carriers' use of other carriers' resources, many technical issues exist in connecting two networks owned by different carriers. A key technical issue is determining where to connect the two networks. This paper introduces a new optimization problem whose objective is to minimize the costs of connecting two networks while improving overall network availability. To solve this NP-hard problem, we propose a deep reinforcement learning algorithm and compare its performance with that of a general greedy algorithm to evaluate the construction cost savings achieved while satisfying the target availability of each network.

In this paper, we discuss the issues of supporting protection switching in multiple recovery domains (MRDs). In MRDs, dual node interconnection (DNI) between two adjacent domains is essential to provide reliable services for customers whose traffic traverses multiple domains. Existing linear protection switching solutions in packet transport networks, such as multiprotocol label switching—transport profile (MPLS-TP) and Ethernet, support protection of an end-to-end flow between two end nodes in a recovery domain; however, they do not provide DNI capability between two recovery domains. We propose a DNI scheme with linear protection switching technology and analyze its advantages. In the case of a large number of connections affected by a single failure in MRDs protected by linear protection, traffic recovery can be delayed owing to the signaling associated with excessive simultaneous signal fail triggers. We design a collective interconnection message (CIM) scheme that improves the performance of linear protection DNI (LP-DNI) in MRDs and confirms that the CIM mechanism reduces restoration time significantly. By extensive simulations on link and node failures scenarios, we investigate different combinations of LP-DNI and existing ring protection processes to establish the influence of service resiliency in MRDs.

In this paper, we discuss the issues of providing protection switching capability for point-to-multipoint (P2MP) connections in both Ethernet and MPLS-TP based packet transport networks. We introduce two types of per-leaf protection that does not require any modification to the existing protection standards: linear protection and ring protection. Although the existing standard schemes can be used in P2MP connection networks, their performances can be improved by a collective signal fail (C-SF) mechanism proposed in this paper. Then, two other schemes, called tree protection and hybrid protection, are newly proposed to reduce the service recovery time when a single failure leads to multiple signal fail events and results in a significant amount of processing burden in a root node. The protocol behavior of tree protection is designed under the objective of a minimal modification to the existing standard protection operation. Hybrid protection scheme is devised to maximize the benefits of per-leaf protection and tree protection. To observe how well each scheme achieves the efficient traffic recovery in a failure event, we evaluate their performance using a test bed with real systems as well as computer simulation based on the formulas driven in this paper.

The MPLS Transport Profile (MPLS-TP) is a framework for the construction and operation of reliable packet-switched transport networks based on the architectures for MPLS and Pseudowires. Its development has been shared between the IETF, where the MPLS expertise resides, and the ITU-T, with its historic understanding of transport networks. MPLS-TP adds two significant features to the MPLS toolkit: Operations, Administration, and Maintenance (OAM), and linear protection switching. Unlike OAM, which resulted in two application specific and incompatible standards being approved at the World Telecommunication Standardization Assembly (WTSA) in November 2012, the MPLS-TP linear protection specification has resulted in a single unified solution that has been published in IETF RFCs and in ITU-T Recommendation G.8131. This article outlines the novel concepts and operation principles of the unified MPLS-TP linear protection switching mechanism and discusses how it differs from pre-existing solutions. In addition, the issues of compatibility with pre-existing solutions and the applicability to other network topologies are discussed.

In connection-oriented data transport services, data loss can occur when a service experiences a problem in its end-to-end path. To resolve the problem promptly, the data transport systems providing the service must quickly modify their internal configurations, which are distributed among different locations within each system. The configurations are modified through a series of problem (event) handling procedures, which are carried out by multiple control processors in the system. This paper proposes a provisioning-to-signaling method for inter-control-processor messaging to improve the time efficiency of event processing. This method simplifies the sharing of the runtime event, and minimizes the time variability caused by the amount of event data, which results in a decrease in the latency time and an increase in the time determinacy when processing global events. The proposed method was tested for an event that required 4,000 internal path changes, and was found to lessen the latency time of global event processing by about 50% compared with the time required for general methods to do the same; in addition, it reduced the impact of the event data on the event processing time to about 30%.

In SDN and NFV technologies, performance of a virtual switch is important to provide network functionalities swiftly. Since the lookup operation of the virtual switch has been considered a major bottleneck of performance, we need to devise an efficient way to reduce the lookup time. To improve the lookup speed, previous research has suggested a compact lookup table in a fast memory, but the issue of collision in a hash table has not been addressed well enough. This paper proposes a new scheme that mitigates collisions by utilizing empty slots of the hash table. We propose to insert a new element that may collide to an empty slot instead of linking it with a pointer. According to an evaluation on our experiments, about 20% of elements that could have experienced collisions in the conventional scheme are inserted into empty slots in each bucket. Moreover, the access time of collided elements has halved by avoiding unnecessary memory accesses.

We propose and verify fault management method using tandem connection monitoring for OTN multiplexing hierarchy through experiments on fiber-cut event. Marking defects on TCM coordinates that consists of TCM levels and ODU trails, and identifying a root-cause of faults using two consequent actions of trail signal fail and server signal fail are newly presented as means to find fault-originators and root-causes. By effectively managing multiple faults all over the network and all over the hierarchical OTN layers, the fault management method could contribute toward the enhanced OAM for fast end-to-end service maintenance and fast restoration.

Self-healing Ethernet rings show promise for realizing the SONET/SDH-grade resilience in Carrier Ethernet infrastructures. However, when a ring is faulty, highpriority protection messages are processed in less time than low-priority data frames are processed. In this situation, any delayed data frames either being queued or traveling through long ring spans will cause the ring nodes to generate incorrect forwarding information. As a result, the data frames spread in the wrong direction, causing the ring to become unstable. To solve this problem, we propose four schemes, that is, dual flush, flush delay timer setting, purge triggering, and priority setting, and evaluate their protection performance under various traffic conditions on a ring based on the Ethernet ring protection (ERP) method. In addition, we develop an absorbing Markov chain model of the ERP protocol to observe how traffic congestion can impact the protection performance of the proposed priority setting scheme. Based on our observations, we propose a more reliable priority setting

Network administrators can make their own programmable network by using an SDN infrastructure with the OpenFlow protocol. Through the OpenFlow protocol, a SDN controller instructs an OpenFlow switch to perform specific actions, such as service chaining, according to header fields of incoming packets. Our main goal is to support an extended view of the OpenFlow architecture by inspecting not only the packet header but also the payload information in the packets. For this purpose, we address a sophisticated packet forwarding scheme using DPI to inspect effectively all incoming packets. According to the results of our experiments, we choose the inside of a virtual switch as the most suitable position of the DPI module. In our scheme, there are a log server, a monitoring application, and DPI function for monitoring and managing network traffic. When the DPI module detects a predefined string pattern of bits in an incoming packet, the switch sends the matching information to the log server that stores logs that contain the detected pattern and resource usage. Then, the monitoring application periodically gathers log information on the log server to compare that information with predefined network policies. Finally, we show that the packets are dealt with in a more effective and efficient way in our sophisticated packet forwarding scheme.

Ethernet ring protection (ERP), defined in ITU-T recommendation G.8032, is a promising technology to realize the SONET/SDH-grade resilience in carrier Ethernet infrastructures. With its own loop prevention technique to block a link called ring-protection-link (RPL) in an ERP ring, the ERP introduces a new optimization problem named RPL positioning whose objective is to find the optimal RPL location to minimize the required ring capacity for given traffic demands. In this paper, we derive the optimal RPL positioning as an integer linear programming (ILP) problem and additionally propose an optimal RPL positioning algorithm for more effective ring loading. By evaluations, we demonstrate that our proposed algorithm is indeed capable of finding the optimal RPL equal to the results by an exhaustive search emerged in prior works or the formulated ILP model. But, it is more efficient in computing time than the exhaustive search and the ILP model.

Ethernet ring protection (ERP) technology, which is defined in ITU-T Recommendation G.8032, has been developed to provide carrier grade recovery for Ethernet ring networks. However, the filtering database (FDB) flush method adopted in the current ERP standard has the drawback of introducing a large amount of transient traffic overshoot caused by flooded Ethernet frames right after protection switching. This traffic overshooting is especially critical when a ring provides services to a large number of clients. According to our experimental results, the traditional FDB flush requires a link capacity about sixteen times greater than the steady state traffic bandwidth. This paper introduces four flush optimization schemes to resolve this issue and investigates how the proposed schemes deal with the transient traffic overshoot on a multi-ring network under failure conditions. With a network simulator, we evaluate the performance of the proposed schemes and compare them to the conventional FDB flush scheme. Among the proposed methods, the extended FDB advertisement method shows the fastest and most stable protection switching performance.

Ethernet technologies are rapidly gaining importance as a prevailing solution for carrier networks. Ethernet ring protection switching, defined in ITU-T Recommendation G.8032, provides a means of achieving carrier network requirements for ring network topologies. This article outlines the novel concepts in this Ethernet ring protection switching mechanism and discusses the fundamental operating principles by which the automatic protection switching (APS) protocol works. In addition, several feature enhancements in protection behavior and solutions being considered for the next phase of standardization are discussed.

In this paper, an algorithm that provides absolute and proportional differentiation of packet delays is proposed with the objective of enhancing quality of service in future packet networks. It features an adaptive scheme that adjusts the target delay for every time slot to compensate the deviation from the target delay, which is caused by prediction error on the traffic to arrive at the next time slot. It predicts the traffic to arrive at the beginning of a time slot and measures the actual arrived traffic at the end of the time slot. The difference between them is utilized by the delay control operation for the next time slot to offset it. Because the proposed algorithm compensates the prediction error continuously, it shows superior adaptability to bursty traffic and exponential traffic. Through simulations we demonstrate that the algorithm meets the quantitative delay bounds and is robust to traffic fluctuation in comparison with the conventional non-adaptive mechanism. The algorithm is implemented with VHDL on a Xilinx Spartan XC3S1500 FPGA, and the performance is verified under the test board based on the XPC860P CPU.

Ethernet technology is rapidly gaining importance as it becomes a dominant solution for a converged transport network. Ethernet OAM features defined in standards provide a means of performance improvement to meet carrier-class transport network requirements. This article outlines Ethernet OAM functions and mechanisms, and explains how its performance monitoring schemes work. In addition, this article introduces open issues and their potential solutions in the performance monitoring of Ethernet OAM for the next phase of standardization.

In this paper, we consider a problem in networks with storage servers for providing multimedia service. The design involves assigning communication link capacity, sizing the multimedia servers, and distributing different types of content at each server, while guaranteeing an upper limit on the individual end-to-end blocking probability. We consider alternative methods for obtaining the end-to-end blocking probability with low computation time and present optimization procedures to obtain an optimal solution. Under a linear cost structure, our numerical investigations consider different scenarios that might be helpful in understanding how to distribute multimedia content for a cost-optimized solution. © 2001 John Wiley & Sons, Inc.

We introduce an improved Ethernet ring protection method, selective filtering database (FDB) advertisement, to minimize traffic overshoot in the event of a failure or recovery. On the protection switching event, the proposed method makes all ring nodes perform an FDB flush except the FDB entries associated with their client subnets. Then, ring nodes rapidly exchange their client MAC address lists so that their FDBs are immediately updated by indirect MAC address learning. The proposed scheme guarantees fast and reliable protection switching over the standard scheme.

Ethernet ring protection is a new technology introduced in ITU-T Recommendation G.8032, which utilizes the generic Ethernet MAC functions. We introduce an alternative enhanced protection switching scheme to suppress penalty in the switching transient, in which the Ethernet MAC filtering database (FDB) is actively and directly modified by information disseminated from the nodes adjacent to failure. The modified FDB at all nodes are guaranteed to be consistent to form a complete new ring network topology immediately. This scheme can reduce the capacity requirement of the G.8032 by several times. This proposed scheme can be also applied in IP protection rings.