Amazon Technologies, Inc.

Geoffrey Scott Pare

Blake Meike

Hok Peng Leung

Chiranjeeb Buragohain

Sumit Lohia

John Walter Heitmann

Christopher David Byskal

Christopher Reddy Kommareddy

Liang Chen

Alex Maclinovsky

Bruce Erwin Brown

Zachary S Musgrave

Reto Kramer

Brian James Schuster

Venkates Paramasivam Balakrishnan

Christopher Whitaker

Daniel Lee Osiecki

Nishanth Shankaran

Christopher Vincent Kaukl

Christopher Paul Dury

Mohanish Hemant Kulkarni

David John Ward, Jr

Joseph Richard Thompson

Alexis Levasseur

Alexander Lambertus Eusman

Sunbir Gill

Matthew Alan Jones

Kiran Kumar Meduri

David Echevarria Ignacio

Antonio Vargas Garcia

David Nunnerley

Rahul Ravikumar

James Andrew Trenton Lipscomb

Brian Helfrich

Peng Zhai

Aleksey A Polesskiy

Kenny Nha Lam

George Elmer Burgyan

Paul Myoungwhan Choi

Xu Wei

Greg Suhyun Kang

Robert Daniel Reid, Jr

Richard Franklin Porter

Zane Grey Reynolds

Daniel Jason Dickinson

Venkates P Balakrishnan

John W Heitmann

Techniques are described for managing program execution capacity, such as for a group of computing nodes that are provided for executing one or more programs for a user. In some situations, dynamic program execution capacity modifications for a computing node group that is in use may be performed periodically or otherwise in a recurrent manner, such as to aggregate multiple modifications that are requested or otherwise determined to be made during a period of time, and with the aggregation of multiple determined modifications being able to be performed in various manners. Modifications may be requested or otherwise determined in various manners, including based on dynamic instructions specified by the user, and on satisfaction of triggers that are previously defined by the user. In some situations, the techniques are used in conjunction with a fee-based program execution service that executes multiple programs on behalf of multiple users of the service.

Dynamically modifying program execution capacity

A resource provisioning service allows users to provision multiple, different network resources in an atomic manner and with a single call to a resource provisioning service. In some instances, the multiple, different network resources comprise individual types of resources that form a portion of one or more cloud-computing platforms. For instance, one or more entities may host and operate a cloud-computing platform that includes different types of network resources, such a storage service, a load balancing service, a compute service, a security service, or any other similar or different type of network-accessible service.

Provisioning multiple network resources

A game-hosting service of a service provider network is configured to place game sessions on fleets of virtual machine (VM) instances using a value-based approach. In order to place a game session on a fleet, the game-hosting service may determine one or more attributes of the game session request, such as player attributes and/or game attributes. The game-hosting service may also determine available fleets of VM instances allocated to a subscriber associated with the game session request, which may be located across disparate geographical regions. A value(s) may be determined based on the attributes of the game session request, and the value(s) may be used to select a fleet of the available fleets for hosting the game session.

Value-based game session placements

A game-streaming service of a service provider network is configured to assign virtual machine (VM) instances to game streaming requests using a value-based approach. In order to assign a VM instance to a game streaming request for streaming a game to a client device of a player associated with the game streaming request, the game-streaming service may determine one or more attributes of the game streaming request, such as player attributes and/or game attributes. The game-streaming service may also determine available VM instances allocated to a subscriber associated with the game streaming request, which may be located across disparate geographical regions. A value(s) may be determined based on the attributes of the game streaming request, and the value(s) may be used to select one of the available VM instances for streaming the game as part of a corresponding game session.

Value-based streaming instance assignment

Live migration can be performed for a process executing on an allocated resource for an application session. During a session it might be determined that it is necessary, or would at least be advantageous, to migrate the process to different resource capacity. New capacity such as a new computing instance can be identified, and synchronized with a current instance to reflect current state information. This can include obtaining state data from a serialized file or transaction log, among other such options. The client devices on the session can be caused to connect or direct state data to the new instance, and when ready the new instance can be responsible for the process, such as by taking over as the application server for the session. Once the client devices are utilizing the new instance, the process can be terminated on the old instance and the instance destroyed or reallocated.

Live migration for hosted sessions

Resources for hosting game sessions can be dynamically selected based upon factors in addition to availability and capacity. A requested session may have various criteria enabling it to be hosted on different types of resources, such as different types of resource instances operating on physical servers. While a fixed amount of capacity can be purchased over an extended period of time, options can also exist that enable purchasing resource capacity for a limited period of time, or purchasing excess capacity without a fixed time limit but that is subject to potential interruption. Benefits relating to cost and performance may offset these potential limitations, and the selection of a type of resource to host a game can be based upon an analysis of these factors. If a game session is to be interrupted or will time out, advance notice can be provided such that the game can take a corresponding action.

Resource selection for hosted game sessions

The present disclosure provides for matching users into user groups based on user attributes. A neural network is trained using training data including user data of users in an observed grouping and various outcomes or optimization parameters associated with the observed grouping. The trained neural network is used to process a set of queried players determine a match quality for the queried players with respect to certain outcomes or optimization parameters based at least in part on the user data associated with the users. The match quality is a measure of how optimized a group containing the queried users is with respect to the optimization parameters and can be compared to a threshold. The queried users may be matched into a group if the match quality meets or exceeds the threshold.

Matching players for networked gaming applications

Techniques for concurrent execution of multiple content item processes, such as video game processes, on a single virtual machine instance are described herein. In some examples, developers may provide an indication of a desired content item process quantity count, which is a quantity of content item processes for concurrent execution of content item sessions on each of one or more virtual machine instances. Also, in some examples, the content item process quantity count may be adjustable such that it may be changed, for example in response to various conditions or events. Also, in some cases, collected performance metric information may be used to make intelligent decisions regarding scaling of content item processes. In some examples, when a content item session stops executing on a healthy content item process, the content item process may remain active and may be reused by executing one or more subsequent content item sessions.

Virtual machine with multiple content processes

A method and apparatus for eviction of session data from a cache are disclosed. In the method and apparatus, an aspect of data of a plurality of aspects of data is associated with a corresponding set of conditions for eviction of a plurality of sets of conditions for eviction from a cache. The set of conditions for eviction corresponding to the aspect of the plurality of aspects of the data are evaluated and the aspect of the data of the plurality of aspects of the data is evicted from cache based at least in part on the evaluation.

Eviction of data in a session service cache

Shared electronic resources can be allocated for a gaming application on a dynamic basis. A package of game content and configuration can be used to allocate an appropriate amount of resource capacity. Task-based instances can be launched as needed and terminated once a gaming session has completed. In order to account for relatively long load times, a number of pre-warmed task-based instances can be allocated that can load the game server functionality up to a determined state. The number of pre-warmed instances can be based on a number of different factors, such as an amount of unused capacity across a shared resource environment. When a request for a session is received, a pre-warmed instance is selected to be upgraded to a game server instance to serve the game session for the request. Pre-warmed instances can be reclaimed if capacity is needed for other purposes.

Dynamic resource allocation for gaming applications

Techniques are described for managing program execution capacity, such as for a group of computing nodes that are provided for executing one or more programs for a user. In some situations, dynamic program execution capacity modifications for a computing node group that is in use may be performed periodically or otherwise in a recurrent manner, such as to aggregate multiple modifications that are requested or otherwise determined to be made during a period of time. In addition, various operations may be performed to attribute causality information or other responsibility for particular program execution capacity modifications that are performed, including by attributing a single event as causing one capacity modification, and a combination of multiple events as possible causes for another capacity modification. The techniques may in some situations be used in conjunction with a fee-based program execution service that executes multiple programs on behalf of multiple users of the service.

Attributing causality to program execution capacity modifications

Techniques for wait time based scaling of content item instances are described herein. In some examples, a customer of a computing service provider may select a wait time threshold for joining one or more of the customer's content items that are hosted by the computing service provider. When a joining user logs-in and requests to join a content item, a calculation may be made regarding an estimated wait time for the user to join one or more existing content item instances. The estimated wait time may then be compared to the wait time threshold. A determination may then be made regarding whether or not to launch a new content item instance in response to the user's join request. In some examples, this determination may be based, at least in part, on the comparison of the estimated wait time to the wait time threshold.

Content item instance scaling based on wait time

Growing community of inventors

Seattle, WA, United States of America

Geoffrey Scott Pare