Amazon Technologies, Inc.

Fabien Chraim

John William Evans

Marina Thottan

Irina Lopatina

Antonio Gomariz Penalver

A metric monitor may obtain data that measures network traffic of a network, through which a host may be accessed by clients, over one or more time windows. The metric monitor may determine one or more metrics based on the data obtained from the network over the time windows, which may represent performance of the network to the host. The metric monitor may provide the metrics to a workload management system to determine placement of a particular workload at the host.

Network performance driven compute workload placement

In computing networks, path availabilities are estimated according to the present disclosure. The path availabilities may be calculated based on connectivity and performance measurements provided by multiple data sources, including passive, active, and/or route monitoring data sources. The measurements may be classified using network topology and processed to determine availability indicators corresponding to the measurements. The availability indicators may be aggregated to determine an overall path availability score for a path associated with the indicators.

Estimating network availability using connectivity and performance data

Packet transmission rate and packet drop rate for discrete network devices in a network are used to estimate end-to-end traffic demand and loss in the network. Data regarding the packet transmission rate and drop rate are passively collected for each network device and transmitted to a network monitoring unit. The network monitoring unit compiles the data and generates a series of simultaneous equations that represent traffic demand and loss between the discrete network devices along the paths connecting respective source-destination pairs. By determining an optimal solution to the simultaneous equations, an estimate of end-to-end traffic loss and corresponding traffic demand, which takes into account packet loss at each network device, can be generated for each source-destination pair. The optimal solution can be formed as a traffic matrix, which aggregates source-to-destination traffic demands, and a loss matrix, which aggregates source-to-destination traffic losses.

Estimating end-to-end network traffic demand and loss

In computing networks, end-to-end loss distributions are estimated according to the present disclosure. The loss distributions may be calculated based on loss values reported by nodes in the network along an end-to-end communication path, which may be identified by a route tracing process. The loss values for each node in the path for a given time window may be reduced by a rounding process, and the reduced values for adjacent nodes in the path may be iteratively multiplied with one another in a pairwise manner to generate an end-to-end loss value vector, where the resulting product from each iteration may be reduced by the rounding process. A graphical or other representation of the values in the end-to-end loss value vector indicating a distribution of losses experienced along the path may be output via a user interface.

Estimating end-to-end loss distributions in computer networks

The numbers of packets transmitted and packets lost for network devices in a computer network are used to estimate a probability of end-to-end packet loss for traffic between a source-destination pair in the network. Metrics of packet transmission and loss at each network device interface are passively collected and transmitted to a network monitoring unit, which uses the metrics to calculate probability of successful packet transmission at each hop. For a particular end-to-end path between the source-destination pair, the network monitoring unit combines the probabilities of successful packet transmission for the hops along the path to yield a probability of successful end-to-end packet transmission. Probability of packet loss along the end-to-end path is determined based on the probability of successful end-to-end packet transmission. Alternatively, the network monitoring unit can estimate which available end-to-end paths between the source-destination pair have a probability of packet loss that satisfies a predetermined loss threshold.

Growing community of inventors

Tucson, AZ, United States of America

Fabien Chraim