China
Africa
Explore chapters and articles related to this topic
High-Performance Switch/Routers
Published in James Aweya, Designing Switch/Routers, 2023
Typically, in high-performance network devices, sFlow and NetFlow can be implemented to provide scalable, wire-speed network monitoring and accounting with no impact on device performance [FOR10ESER05]. sFlow or NetFlow can be integrated into the forwarding capabilities of the device to collect and aggregate details on traffic flows at different layers of the OSI reference model (from Layer 2 through Layer 4) and automatically deliver that information to a network management station. The network management station may then employ, for example, a Java-based network configuration and management tool to display, in graphical detail, network and application-level traffic information.
Avaya P580 and P882 Routing Switch Architecture with 80-Series Media Module
Published in James Aweya, Switch/Router Architectures, 2019
Using a forwarding engine directly with a full topology based forwarding table (not a route/flow cache) provides a much more powerful and efficient architecture for the design of networks that aim to deliver improved availability, performance, and scalability. Such networks also enable important services that include: Multiprotocol Label Switching (MPLS) with sophisticated network traffic engineering that allows for the creation of services such IP Virtual Private Networks (VPNs)Network monitoring and collection tools such as NetFlow and sFlow [RFC3176] that allow for gathering network statistics. sFlow and NetFlow are tools that can be used to generate detailed information on traffic flows in a network to help network operators analyze their traffic patterns and accurately plan network capacity.Quality of service (QoS) functions such as traffic policing, traffic shaping, WFQ, WRED and other traffic management mechanisms that help prevent one application (particularly, one generating best-effort traffic) from hogging network bandwidth and starving out other applications (particularly, ones generating real-time traffic).The approach of using a forwarding engine with both optimized lookup mechanism and topology based forwarding table (referred to as Cisco Express Forwarding (CEF) by Cisco Systems) can be implemented in a distributed architecture where processing tasks are spread across the line cards—distributed forwarding engines (with associated forwarding tables). In the distributed forwarding architecture, the line cards maintain an identical copy of the forwarding table and adjacency information. The line cards perform local forwarding of packets, thus relieving the route processor (control plane) of direct involvement in the forwarding process.
Low cost network traffic measurement and fast recovery via redundant row subspace-based matrix completion
Published in Connection Science, 2023
Kai Jin, Kun Xie, Jiazheng Tian, Wei Liang, Jigang Wen
The traffic flow exchanged between OD pairs can be measured using flow-based measurement tools like NetFlow (Claise, 2004) and sFlow (Wang et al., 2004). However, because these tools frequently consume too many CPU and memory resources (Yu et al., 2013), obtaining the TMs by measuring the traffic volumes between all OD pairs at all time slots is impractical due to high measurement overhead (Li, Liang et al., 2023; Xie et al., 2017, 2021, 2022). To reduce measurement overhead, matrix completion-based sparse network monitoring attracted much recent interest (Li et al., 2023b; Xie, Wang et al., 2017). Sparse network monitoring aims to get the entire TM data by taking measurements in only a portion of OD pairs and inferring the un-measured data through matrix completion. According to matrix completion theory, if the target matrix has a low-rank feature, it can be accurately recovered by a subset of samples. Some studies (Roughan et al., 2011; Xie et al., 2021) have indicated that TMs have a temporal and spatial correlation, implying that TMs meet the low-rank prerequisite for using matrix completion. Although promising, existing matrix completion-based sparse network monitoring suffers from two major problems: Low recovery accuracy: In addition to the low-rank feature, matrix coherence is another essential feature in the matrix completion theory. When the target matrix is highly coherent, which implies that not all entries reveal the same information, most of the information will be omitted if we take samples randomly. Using these random samples to recover the target matrix will lead to poor recovery accuracy. Some recent studies (Roughan et al., 2011) have applied matrix completion to infer un-measured data by some random measurement samples. However, these samples may not load enough information to recover unmeasured data accurately.Long processing time: To achieve better recovery accuracy, some studies (Liu & Wang, 2020; Xie et al., 2015) aim to select better measurement sampling locations following the matrix completion framework. However, to select a better measurement sampling location, such methods usually require multi-round iterations, which results in high computational costs. Besides, existing matrix completion algorithms (e.g. SRMF Roughan et al., 2011, SVT Cai et al., 2010, ALM Lin et al., 2010) involve an iterative training parameters process, which incurs high computational costs too.