Showing papers by "Arista Networks published in 2018"

Toward transport ecosystem interoperability enabled by vendor-diverse coherent optical sources over an open line system

Abstract: Optical line-side interoperability is a critical component for enabling cloud service providers to deploy capacity at scale with a diverse supply chain. Historically, vendor-locked line systems and proprietary coherent digital signal processing (DSP) technologies have been barriers to this goal. Advances in open line platforms and coherent interfaces embedded directly in layer-2/3 switches and routers have begun paving the way toward full end-to-end interoperation. In this paper, we take a partial step toward demonstrating end-to-end interop by interoperating vendor-diverse packet switching platforms from Arista, Cisco, and Juniper, which utilize the same coherent DSP chip set, over an open line system (OLS). All three vendors' switches feature integrated coherent optics and are shown to interoperate over 2000 km at 150G 8QAM and 1000 km at 200G 16QAM on Microsoft's OLS.

Morpheus: Exploring the Potential of Near-Data Processing for Creating Application Objects in Heterogeneous Computing

Abstract: In modern computing systems, object deserialization can become a surprisingly important bottleneck-in our test, a set of generalpurpose, highly parallelized applications spends 64% of total execution time deserializing data into objects This paper presents the Morpheus model, which allows applications to move such computations to a storage device and bypass the overhead on the host system We use this model to deserialize data into application objects inside storage devices, rather than in the host CPU Using the Morpheus model for object deserialization avoids unnecessary system overheads, frees up scarce CPU and main memory resources for compute-intensive workloads, saves I/O bandwidth, and reduces power consumption In heterogeneous, coprocessor- equipped systems, Morpheus allows application objects to be sent directly from a storage device to a co-processor (eg, a GPU) by peer-to-peer transfer, further improving application performance as well as reducing the CPU and main memory utilizations This paper implements Morpheus-SSD, an SSD supporting the Morpheus model Morpheus-SSD improves the performance of object deserialization by 166x, reduces power consumption by 7%, uses 42% less energy, and speeds up the total execution time by 132x By using NVMe-P2P that realizes peer-to-peer communication between Morpheus-SSD and a GPU, Morpheus-SSD can speed up the total execution time by 139x in a heterogeneous computing platform

System and method of network operating system containers

Abstract: A method and apparatus of a network element that processes control plane data in a network element is described. In an exemplary embodiment, the device receives control plane data with a network element operating system, where at least a functionality of the network element operating system is executing in a container of the network element. In addition, the network element includes a data plane with a plurality of hardware tables and the host operating system. Furthermore, the network element processes the control plane data with the network element operating system. The network element additionally updates at least one of the plurality of hardware tables with the process control plane data using the network element operating system.

Analog Coherent Optics for Long Haul Datacenter Regional Networks

Abstract: We review key aspects of analog coherent optics adoption in long haul datacenter regional networks. We discuss the main advantages and drawbacks of this approach and present potential improvements for next generation 60 + Gbaud long haul analog 8/16QAM systems. Finally, specificities of the 120 km datacenter interconnect (DCI) use case are also considered in the context of the ongoing OIF 400ZR standard effort.

Method and system for processing spanning tree protocol (STP) state in a multichassis link aggregation (MLAG) domain

Abstract: A method and system for processing spanning tree protocol (STP) state in a multichassis link aggregation (MLAG) domain. More specifically, a pair of MLAG peers in the MLAG domain receive and transmit bridge protocol data units (BPDUs) in order to update a spanning tree used to implement the STP. The pair of MLAG peers includes an active MLAG peer and a standby MLAG peer operatively connected via a MLAG peer link. Further, processing of STP state towards updating the spanning tree is held accountable by the active MLAG peer until it experiences failover, after which responsibility for processing the STP state transfers to the standby MLAG peer. Duplicate BPDUs received by the MLAG domain are tracked to refresh a retention period for the STP state retained by the MLAG domain, and thus, prevent the premature discarding of STP state.

Heat sink for optical transceiver

Abstract: An electronics module is provided. The electronics module includes a housing at least partially enclosing a first printed circuit board configured to couple the electronics module to a connector attached to a second printed circuit board. The electronics module includes a first heat sink disposed along a first surface of the housing and a second heat sink disposed along a second surface of the housing. One or more notches or apertures of the first printed circuit board are proximate to the connector thereby enabling an airflow through the second heat sink along the second surface of the housing to exhaust over a surface of the connector with an airflow through the first heat sink.

System and method of dynamic hardware policer allocation

Abstract: A method and apparatus of a network element that installs a control plane data rule from system memory to specialized hardware memory in a network element is described. In an exemplary embodiment, the network element receives control plane data processing statistics of control plane data that is received by the network element, where the control plane data is policed using a plurality of control plane data policing rules stored in the system memory. The network element further determines that a candidate policing rule is being utilized greater than a threshold based on at least the control plane data processing statistics, where the candidate policing rule is one of the plurality of control plane data policing rules stored in the system memory. The network element additionally installs the candidate policing rule in the specialized hardware memory, where the candidate policing rule is subsequently used by the network element to police the control plane data.

Simulating a topology of network elements

Abstract: A method and apparatus of a device that simulates a plurality of network elements is described. In an exemplary embodiment, the device receives network topology information for the plurality of simulated network elements. The device further instantiates a container for each of the plurality of simulated network elements. The device additionally configures a set of processes for each of the plurality of containers, where each of the set of processes simulates at least one of the plurality of simulated network elements. The plurality of set of processes further implements a network topology represented by the network topology information. The device performs a test of the network topology and saves the results of the test.

Cluster file replication

Abstract: A cluster file replication system is provided. Each controller of the plurality of controllers is configured to access a filesystem having a plurality of files including a system database of a controller having state information of the plurality of controllers. Each controller is further configured to have one or more service agents. The one or more service agents of each controller is configured to respond to one of the plurality of controllers becoming a master controller of the cluster. The one or more service agents of each controller is configured to set up one or more objects that react to the state information and coordinate replication of changes to the files, system database and state information from the master controller to follower controllers in the cluster. The one or more objects on each of the follower controllers is supportive of the follower controllers receiving the changes but disabled from initiating the replication. A method for cluster file replication is also provided.

MobiCom'18 Panel: Hammer & Nail vis-a-vis AI / ML Applications to Networked Systems

Abstract: Artificial Intelligence (AI) and Machine Learning (ML) approaches, well known from IT disciplines, are beginning to excite the networking and networked systems community. Of late, we are seeing a huge excitement about applying AI and ML to networked systems. Is this merely a hype? Are there use cases and genuine applications that could lead to real deployment and practical solutions? What are the key challenges in applying AI and ML to networked systems? Can researchers and practitioners in communication networks and networked systems tap into machine learning and AI techniques to optimize network architecture, control and management, leading to increased automation in network operations? Can researchers and practitioners in the AI community explore synergy with networking researchers to optimize network architecture and design? The above are some of the questions that would be addressed during the panel discussion. The objective of the panel discussion would be to tap the minds of the global experts in order to understand the merits and limitations and the future landscape in the intersection of networking/networked systems and AI/ML.

Method and apparatus to mitigate assembly torsion

Abstract: A mechanism to mitigate assembly torsion on an electronics assembly is provided The mechanism including an electronics assembly and a first connector, mounted to the electronics assembly with a lower portion of the first connector proximal to the electronics assembly and an upper portion of the first connector distal to the electronics assembly The mechanism includes a spring, mounted so as to press the upper portion of the first connector and preload the first connector against assembly force imparted by assembly of the first connector to a second connector A method to mitigate assembly torsion on an electronics assembly is also provided

Low-latency network switching device with latency identification and diagnostics

Abstract: A network switch device includes an L1 switch having a first set of external ports and a first set of internal ports. The network switch device further includes an L2+L3 switch having a second set of internal ports, the L2+L3 switch operatively coupled to the L1 switch via the first set of internal ports and the second set of internal ports.

EMI absorber ring

Abstract: An electromagnetic interference absorber for an integrated circuit is provided. The absorber includes a geometric ring of electromagnetic energy absorbing material, dimensioned to fit over a ball grid array (BGA) integrated circuit package assembled to a substrate. The geometric ring has at least one projection arranged to fit into a gap between the substrate and a body of the BGA integrated circuit package so as to retain the geometric ring to the BGA integrated circuit package. Methods to contain electromagnetic interference and to manufacture an electromagnetic interference absorber are also provided.

Hitless upgrades of containers

Abstract: A method and apparatus of a network element that hitlessly upgrades a network element operating system of a network element is described. In an exemplary embodiment, the network element receives a second image for the network element operating system, where a first image of the network element operating system is executing as a first set of processes in a first container and the first set of processes manages the plurality of hardware tables for the network element. The network element further instantiates a second container for the second image. In addition, the network element starts a second set of processes using at least the second image in the second container. The network element additionally synchronizes state data between the first set of processes and the second set of processes. Furthermore, the network element sets the second set of processes as managing the plurality of hardware tables, and stops the first set of processes within the first container.

Method and system for centralized controller for audio visual broadcasts

Abstract: In general, embodiments of the invention relate to a domain controller for a network. The domain controller includes functionality to manage the provisioning of paths in the network in order to enable content to be streamed across the network.

System and method of a centralized gateway that coordinates between multiple external controllers without explicit awareness

Abstract: A method and apparatus of a device that restricts data communicated between a plurality of computing management devices and a plurality of computing devices is described. In one embodiment, a gateway receives network connectivity information of a network for a plurality of computing management devices and a plurality of computing devices, where the plurality of computing devices is coupled to the plurality of computing management devices via a plurality of network elements. In addition, the gateway determines a plurality of sub-networks using at least the network connectivity information, wherein for each of the plurality of sub-networks, at least one of the plurality of computing management devices and a subset of the plurality of computing devices is associated with that sub-network. Furthermore, for each of the plurality of sub-networks and for each pair of computing devices in the associated subset of plurality of computing devices, the gateway determines a policy for data being communicated between this pair of computing devices and the gateway configures at least one of the plurality of network elements coupled to this pair of computing devices to allow data communicated between this pair of computing devices using at least the first policy.

Method for self-calibration of an electrical and/or optical channel

Abstract: An auto-equalization network device for optical transmitting and receiving is provided. The device includes a network device having an optical transmitter and an optical receiver. The network device is configured to determine a frequency domain forward transmission loss characterization for a signal transmitted by the optical transmitter, looped back and received by the optical receiver. The device is configured to generate a model in the frequency domain having control points based on the characterization, generate and load finite impulse response (FIR) filter taps into a finite impulse response filter based on the model in the frequency domain. The device is configured to iterate transmission of a signal with frequency-dependent preemphasis by the finite impulse response filter, characterization for the frequency-dependent preemphasized signal as looped back and received by the optical receiver, comparison to previous characterization, adjustment of the control points, and reloading the finite impulse response filter taps, until the comparison meets an optimum, so that the network device is auto-equalized for optical transmitting.

Updating interface list of multi-link group

Abstract: A method and apparatus of a network element that updates an interface list of a multi-link group of a network element is described. In an exemplary embodiment, a network element receives an indication that the interface list for the multi-link group on a network element is to change. In addition, the interface list includes a first set of interfaces. The network element further includes an address table having a plurality of address entries, where each of the plurality of address entries includes an address, a multi-link nexthop, and a tag group reference. The tag group reference references an entry in a tag group table, where a tag group entry includes a tag group identifier and a tag group set of interfaces. The network element further receives a second set of interfaces. The network element additionally updates the interface list for the tag group entry to include a second set of interfaces. Furthermore, the network element transmits data with the multi-link group using the second set of interfaces.

System and method of dynamic establishment of virtual private networks using border gateway protocol ethernet virtual private networks technology

Abstract: A method and apparatus of a network element that dynamically establishes a first virtual private network (VPN) tunnel is described. In an exemplary embodiment, the network element detects data destined for a first private subnet. In response to the detecting, the network element determines that a next hop for the data does not have an established VPN tunnel that allows access to the first private subnet. The network element further establishes the VPN tunnel and sends the data using the VPN tunnel.

Method and system for managing workloads in a cluster

Abstract: In general, embodiments of the invention relate to a switch that includes a processor, ports, and memory that includes instructions, which when executed by the processor perform a method. The method includes obtaining, via a port, a global task-server mapping from a first server, detecting that one of the ports is congested. The method further includes, based on the detecting, identifying a task associated with the one of the ports using the global-task server mapping, generating a task listing based on the identifying, generating an alert including the task listing, and transmitting the alert to an administrator.

System and method of a cloud service provider virtual machine recovery

Abstract: A method and apparatus of a device that recovers accessibility for an inaccessible virtual machine hosted by a cloud service provider is described. In an exemplary embodiment, the device receives an indication that a recovery disk has been attached to a virtual machine hosted by a cloud service provider, wherein the virtual machine is inaccessible to a client. In addition, the device executes an agent that recovers the accessibility of the virtual machine for the client.

On the Use of a Measurement Correlation Service for Measurement Federations

Abstract: The diversity of services that operate in the Internet has increased significantly in the last years. Performance problems in these services cause important financial losses. To ensure that these problems do not occur, service levels need to be monitored. One of the main techniques for such monitoring involves the use of active measurement mechanisms. However, these mechanisms are expensive in terms of resources consumption due to the activation of measurement sessions. Measurement sessions usually cover only a fraction of what could be measured, which leads to service level problems being missed. Measurement federations can help network administrators in different tasks, such as controlling the activation of active measurement sessions. In this context, measurement correlation can be deployed in order to improve this control in such federations. The main contribution of the present work is the proposal of a data transformation service that provides measurement correlation. This service is used to enable cooperation features in measurement federations, while decreasing resource consumption. Besides that, statistical tests that can be used to compose such correlation are presented. The proposed solution is evaluated using an active measurement dataset from the Brazilian National Research and Education Network (Rede Nacional de Ensino e Pesquisa - RNP). Our results provide insights regarding measurement correlation from federated measurement points and can be used for the design of better application to control active measurement sessions.

Dynamically installing a device driver by a network element

Abstract: A method and apparatus of a network element that installs a device driver used to manage hardware of the network element is described. In an exemplary embodiment, the network element detects, with a functionality of a network element operating system, the hardware of a data plane of the network element, where at least one component of the network element operating system is executing in a first container as a first set of processes. The network element further determines a device driver for the hardware and installs the device driver in a kernel of the host operating system. The network element additionally manages the data, with the network element operating system, using the device driver.