Amazon Technologies, Inc.

Jonah Beckford

Yohan Mammen

Shi Yao Zhang

The use of linear programming may enable the achievement of real-time task execution prioritization. A linear model may be constructed based at least on a set of computing tasks, a linearly expressible goal for the set of computing tasks, one or more linear resource constraints, and one or more computing task dependencies. The linear model is then used to calculate a shadow price for each of a plurality of resource time prediction constraints, in which each shadow price may represent a priority value for a corresponding computing task. When a computing resource becomes available, a computing task with a highest priority value that is able to use the computing resource may be executed.

Resource constrained task scheduling

Disclosed are various embodiments for the management of dependencies on data. Multiple jobs are executed in one or more computing devices. It is determined that a first one of the jobs consumes data produced by a second one of the jobs, and a dependency is recorded. It is then determined whether the dependency is satisfied. When the dependency is determined to be unsatisfied, execution of the first one of the jobs is suspended. When the second one of the jobs produces the data, the dependency is updated as being satisfied. Execution of the first one of the jobs is resumed in response to the dependency being updated as being satisfied.

Managing data dependencies among multiple jobs using separate tables that store job results and dependency satisfaction

Disclosed are various embodiments for managing data indexed by a search engine. A respective identifier from a set of identifiers is assigned randomly in one or more computing devices to each one of a plurality of data records. A set of search queries is provided. A first one of the search queries encompasses all of the set of identifiers. A second search index is generated from a first search index of the data records. The first search index is queried for at most a predetermined number of unmarked ones of the data records using a search query. Each one of a result set of data records that is thereby obtained is inserted into the second search index. The result set of data records is marked in the first search index.

Growing community of inventors

Bothell, WA, United States of America

Jonah Beckford