Versa Networks, Inc.

Twitter, Inc.

Sridhar Vishwanathan Iyer

Apurva Mehta

Akshay Adhikari

Amey Gavand

Aswin Sivasubramanian

Sandeep Pandey

Yue Lu

Pengjun Pei

Dave Traugott

Ziqi Huang

Suraj Chandrasekaran

Xiaojiang Guo

Van Lam

Yong Wang

Zhiyong Xie

Lanbo Zhang

Gary Lam

Bingjun Sun

Qi Li

Shuanghong Yang

Dong Wang

Embodiments of methods and systems for managing traffic across a WAN are disclosed. A method for managing traffic across a WAN involves identifying incoming traffic as Direct Internet Access (DIA) traffic at a first node, identifying incoming traffic as SD-WAN traffic at the first node, policing the DIA traffic at the first node, and adapting the policing of the DIA traffic at the first node in response to a comparison between the SD-WAN traffic and an SD-WAN traffic threshold.

Methods and system for adaptively managing the policing of network traffic

Embodiments of methods and systems for managing traffic across a WAN are disclosed. A method involves advertising an input rate limit across a WAN from a first node, advertising an input distribution array across the WAN from the first node, wherein the input distribution array includes forwarding class-specific weights, and adapting at least one of the input rate limit and the input distribution array in response to an error between a target forwarding class distribution and an observed input traffic distribution.

Methods and system for adaptively managing the distribution of network traffic

An example system includes prediction workers, training workers, and a parameter server. The prediction workers store a local copy of a machine-learned model and run the mode exclusively in serving mode. The training workers store a local copy of a machine-learned model and a local snapshot and run the local copy exclusively in training mode and compare the local model or state to the snapshot after training to send delta updates to the parameter server after training. The parameter server aggregates received delta updates into a master copy of the model, sends the aggregated updates back to training workers and provides two types of updates; a real-time update based on a comparison of the master model with a local snapshot, and a full update. The real-time update occurs at least an order of magnitude more frequently than the full update and includes a subset of the weights in the model.

Efficient online training for machine learning

A method for shaping traffic from an egress port in a software-defined wide-area network (SD-WAN) involves obtaining a stored network bandwidth measurement of the network bandwidth between a source endpoint and a destination endpoint, obtaining a current shaping rate used by the source endpoint when sending data to the destination endpoint, obtaining an updated measurement of the network bandwidth between the source endpoint and the destination endpoint, determining a new shaping rate based on the stored network bandwidth measurement, the current shaping rate measurement, and the updated measurement of network bandwidth, and configuring the shaping rate used by the source endpoint when sending data to the destination endpoint with the new shaping rate.

Method and system for shaping traffic from an egress port in a software-defined wide area network

Embodiments herein disclose methods for selecting one or more paths for routing application traffic based on application performance metrics for hosted applications. In an embodiment, to select the best path, the available paths can be monitored to understand the performance of specific applications through those paths. Subsequently, the performance data is used to derive an application link score (ALS) for any given combination of application and path. The ALS is then be used to determine the best path for a given application.

Application performance based path-selection

Embodiments herein disclose methods for selecting one or more paths for routing application traffic based on application performance metrics for hosted applications. In an embodiment, to select the best path, the available paths can be monitored to understand the performance of specific applications through those paths. Subsequently, the performance data is used to derive an application link score for any given combination of application and path, wherein the application link score is generated as a function of packet delay and packet loss. The ALS is then be used to determine the best path for a given application.

Application performance based path-selection using dynamic metrics

A method involves setting a link aggregation control protocol (LACP) link state for all links in a first service node to STANDBY to put at least one multiplexer in the first service node in a WAITING state to disable frame collection at the redundant service node and setting the LACP link state for all links in a second service node to SELECTED to put at least one multiplexer in the second service node in a COLLECTING/DISTRIBUTING state to enable frame collection at the primary service node.

Method for configuring a connection using redundant service nodes

A method for shaping traffic from an egress port in a software-defined wide-area network (SD-WAN) involves obtaining a stored network bandwidth measurement of the network bandwidth between a source endpoint and a destination endpoint, obtaining a current shaping rate used by the source endpoint when sending data to the destination endpoint, obtaining an updated measurement of the network bandwidth between the source endpoint and the destination endpoint, wherein the updated measurement of network bandwidth is obtained from an SD-WAN controller, determining a new shaping rate based on the stored network bandwidth measurement, the current shaping rate measurement, and the updated measurement of network bandwidth, and configuring the shaping rate used by the source endpoint when sending data to the destination endpoint with the new shaping rate.

A method for shaping traffic across a wide area network is disclosed. The method involves advertising a data rate limit across a wide area network (WAN) from a first node, measuring the rate of data received at a WAN interface of the first node, and, if the measured rate of data received at the WAN interface of the first node exceeds a maximum threshold, advertising a reduced data rate limit across the WAN, and, if the measured rate of data received at the WAN interface of the first node is below a minimum threshold, advertising an increased data rate limit across the WAN.

Method and computing device for shaping traffic across a wide area network

A first and second thread-local counter is allocated to a first and second thread respectively, where the first thread-local counter is updatable only by or on behalf of the first thread and the second thread-local counter is updatable only by or on behalf of the second thread. The first and second thread-local counter are updated and the updated values are communicated to a central process. The central process updates a single counter in a central database by a value equal to the sum of the updated counter values.

Growing community of inventors

Fremont, CA, United States of America

Sridhar Vishwanathan Iyer