Amazon Technologies, Inc.

Paul Derek DeMarco

Terence Smith

Christopher Keakini Kaulia

Michael Shane Thyssen

Chaitanya Avinash Kulkarni

Jeremy Boynes

Christopher Allen Suver

Mark Noel Kelly

Michael James McInerny

Anton Vladilenovich Goldberg

Melissa Elaine Davis

Alexander Julian Tribble

Zachary Jared Wiggins

Charles Alexander Carman

David Y Yamanoha

Paul Walker Mooring

Rohit Rangan

A first node can be selected as a target node, and a second node can be selected as a control to determine a performance limit associated with providing a service. A request for the service can be determined to be routed to the first node. A first indication of a first performance measurement associated with processing a first load by the first node can be received. A second indication of a second performance measurement associated with processing a second load by the second node can be received. A performance error of the first node can be determined based on the first performance measurement and the second performance measurement. The performance limit can be determined based on the performance error, and an action can be triggered based on the performance limit.

Real-time load limit measurement for a plurality of nodes

Request routing in a network of computing devices. Back-end nodes associated with a request router send load reports to request routers. To select a back-end node for processing a request, a request router determines, based on the reports, estimated loads and corresponding errors of the estimated loads for respective nodes. A load balancing technique may be selected, based on the estimated loads and corresponding errors (e.g., round robin may be selected if the significance of an average error outweighs the difference between estimated load values for the nodes, or least estimated load may be selected of the difference between load values is greater than the average error). The selected technique is applied to choose one of the nodes to process the request. Load probability distribution may also be used for request routing, in combination with or separate from the selection based on the estimated loads.

Load balancing technique selection based on estimated load

Request routing in a network of computing devices. Back-end nodes associated with a request router send load reports to the request router that makes routing determinations for request from clients based on the reports. To select a back-end node for processing a request, a request router determines, based on the reports and one or more factors, load probability distributions for each of a group of back-end nodes selected from available back-end nodes. A load value is then randomly selected for each back-end node in the group, based on the node's probability distribution. The node in the group with the least randomly-selected value is chosen to process the request.

Load balancing based on randomized selection of a load value based on a load probability distribution

A system for transmitting data over a network may include data processed according to a data-redundancy encoding technique such as erasure coding to be transmitted via an unreliable, connectionless transmission protocol, for example user datagram protocol ('UDP'). A transmitting manager may receive a request from an application to transmit data. The transmission manager may select encoding parameters and encode data to be transferred using a data-redundancy encoding technique. The transmission manager may initiate transmission of the encoded data and payload header data may via a network, for example via a network communication stream or protocol stack. A reconstruction manager associated with the receiving node may reconstruct the original data using the received erasure coding data and payload header data. In some embodiments, the receiving node may transmit telemetry data to the erasure coding algorithm. The erasure coding algorithm may use received telemetry data to adjust parameters of the erasure coding algorithm for future erasure coding and transmission operations. In some systems, telemetry data may include a request to retransmit all or a portion of the erasure coded data.

Transfer of data-redundancy encoded data via unreliable, connectionless protocol

Systems and methods are disclosed to implement a gossip protocol to synchronize data in among nodes of a distributed computing system. During a round of the gossip protocol, a first node generates a space-efficient probabilistic data structure (SEPDS) from its data set. The SEPDS is generated using a set of hash functions that changes from round to round. The set of hash functions may be derived using two base hash functions without reliance on the use of any randomizing operations, and the result of each hash function may be assigned to modify a different portion of the SEPDS. The generated SEPDS is sent to a second node, which performs probabilistic queries on the SEPDS to compare the contents of its own data set with the SEPDS. Any data items that are missing from the SEPDS are sent back to the first node, which updates its data set accordingly.

Generating probalistic data structures in gossip protocols

Described are techniques for determining the effect of a change in how data is handled by a first device on other devices within a system. At a first device, sending, receiving, or processing of data may be prevented for a length of time. Metric data may be determined from other devices during the length of time and during one or more time periods before or after the length of time. The metric data may indicate the effect of the handling of data by a first device on the one or more other devices.

System for modeling distributed systems

Various examples are directed to optimizing communications based on network conditions. A computing device may be programmed to receive from a process executing at the computing device a network transmission request. The network transmission request may indicate payload data and a recipient computing device. The computing device may transmit to a recipient computing device a plurality of data packets. Each of the plurality of data packets may comprise at least a portion of the payload data. The computing device may detect an improvement in network conditions and accelerate retransmission of at least one data packet.

Growing community of inventors

Seattle, WA, United States of America

Paul Derek DeMarco