Amazon Technologies, Inc.

Other

Tyson James Lamoreaux

Michael David Marr

Jagwinder Singh Brar

Mark Y Kelly

Justin Oliver Pietsch

Alan Michael Judge

Mark Noel Kelly

Daniel Todd Cohn

Karl Andre McCabe

Carlos Vara Callau

Timothy Richard Kohn

Sonali Roy

Eoin Francis Cavanagh

Alan M Judge

Edward Crabbe

Daril Alexandre Vilhena

Brian Myungjin Choi

Generally described, systems and methods are provided for detecting the impact of network failures. The system collects performance information from a plurality of nodes and links in a network, aggregates the collected performance information across paths in the network, processes the aggregated performance information for detecting failures on the paths, adjusts the set of performance information by removing the performance information for any nodes considered to be associated with performance information that is statistically different from performance information from other nodes at a given location or extrapolates the collected information to other paths, and determines the impact to customers of the network failures detected using the adjusted set of performance information.

Determining impact of network failures

Techniques for dynamic augmentation of server rack network capacity are provided herein. Network nodes are described that are connected between server rack switches and a rack aggregation layer device. A port allocation controller is also described that can automatically reconfigure the network nodes to create connections, via the network nodes, between the rack aggregation layer device and the server rack switches. These network node connections can be used to alter the network capacities of the server rack switches in response to changing network demands.

Dynamic network capacity augmentation for server rack connectivity

Efficient and highly-scalable network solutions are provided that each utilize deployment units based on Clos networks, but in an environment such as a data center of Internet Protocol-based network. Each of the deployment units can include multiple stages of devices, where connections between devices are only made between stages and the deployment units are highly connected. In some embodiments, the level of connectivity between two stages can be reduced, providing available connections to add edge switches and additional host connections while keeping the same number of between-tier connections. In some embodiments, where deployment units (or other network groups) can be used at different levels to connect other deployment units, the edges of the deployment units can be fused to reduce the number of devices per host connection.

Connecting network deployment units

Host machines and other devices performing synchronized operations can be dispersed across multiple racks in a data center to provide additional buffer capacity and to reduce the likelihood of congestion. The level of dispersion can depend on factors such as the level of oversubscription, as it can be undesirable in a highly connected network to push excessive host traffic into the aggregation fabric. As oversubscription levels increase, the amount of dispersion can be reduced and two or more host machines can be clustered on a given rack, or otherwise connected through the same edge switch. By clustering a portion of the machines, some of the host traffic can be redirected by the respective edge switch without entering the aggregation fabric. When provisioning hosts for a customer, application, or synchronized operation, for example, the levels of clustering and dispersion can be balanced to minimize the likelihood for congestion throughout the network.

Clustered dispersion of resource use in shared computing environments

Fused data center fabrics

Host machines and other devices performing synchronized operations can be dispersed across multiple racks in a data center to provide additional buffer capacity and to reduce the likelihood of congestion. The level of dispersion can depend on factors such as the level of oversubscription, as it can be undesirable in a highly connected network to push excessive host traffic into the aggregation fabric. As oversubscription levels increase, the amount of dispersion can he reduced and two or more host machines can be clustered on a given rack, or otherwise connected through the same edge switch. By clustering a portion of the machines, some of the host traffic can he redirected by the respective edge switch without entering the aggregation fabric. When provisioning hosts for a customer, application, or synchronized operation, for example, the levels of clustering and dispersion can be balanced to minimize the likelihood for congestion throughout the network.

Efficient highly connected deployment units

The deployment and scaling of a network of electronic devices can be improved by utilizing one or more network transpose boxes. Each transpose box can include a number of connectors and a meshing useful for implementing a specific network topology. When connecting devices of different tiers in the network, each device need only be connected to at least one of the connectors on the transpose box. The meshing of the transpose box can cause each device to be connected to any or all of the devices in the other tier as dictated by the network topology. When changing network topologies or scaling the network, additional devices can be added to available connectors on an existing transpose box, or new or additional transpose boxes can be deployed in order to handle the change with minimal cabling effort.

Transpose box based network scaling

The efficiency of scaling of a network of computing devices can be improved by grouping highly-connected portions into deployment units wherein devices only have to be aware of other devices in the respective deployment unit. The various deployment units can be connected by a backbone mechanism. In some embodiments, the backbone utilizes a torroidal connection scheme to connect the backbone switches and pass routing protocol information. Host traffic can still utilize connections between tiers, such that standard routing and networking protocols can be utilized. In other embodiments, logical interfaces can be initialized in each backbone switch, such that each switch itself can function as a point backbone. The devices of the various deployment units then can communicate with each other via the point backbones using standard networking and routing protocols.

Point backbones for network deployment

Transpose boxes for network interconnection

Growing community of inventors

Seattle, WA, United States of America

Tyson James Lamoreaux