Amazon Technologies, Inc.

Marc Levy

Marc John Brooker

Norbert Paul Kusters

John Robert Smiley

Bei-Jing Guo

Madhuvanesh Parthasarathy

Christopher Magee Greenwood

Surya Prakash Dhoolam

Tarun Goyal

David R Richardson

Danny Wei

Marc Stephen Olson

Gary Michael Herndon

Joshua Dawie Mentz

Tao Chen

Michael Groenewald

Steven J Kruy

Mustafa I Abrar

Mark Danoher

Gabriël De Vos Smit

Burst throttling methods may be used to manage computing resources of a data storage service. Tokens may represent I/O operations executed by a customer of the data storage service. A first token bucket may contain a set of tokens representing the overall I/O operation capacity of the data storage service. Additionally, a second token bucket may contain a set of tokens for a given logical volume maintained by the data storage service. When I/O requests are received tokens may be charged the first token bucket and the second token bucket. Furthermore, if there is sufficient capacity, the data storage service may charge a reduced number of tokens to the third token bucket.

Burst throttling for multi-tenant storage services

A customer of a computing resource service provider submits a request to a virtual computer system service to instantiate a virtual computer system. In response to the request, the service obtains physical network information in order to identify any hardware groupings that may be used, at a later time, to provision one or more data storage volumes that may be associated with the virtual computer system. The service assigns a score to each of these hardware groupings based at least in part on the available capacity for provisioning of these data storage volumes. Based at least in part on these assigned scores, the virtual computer system service selects a hardware grouping and instantiates the virtual computer system within the selected hardware grouping.

Hardware grouping based computing resource allocation

Computing resource services have computing resources allocated among regions and data zones, where regions can be completely separate deployments of an infrastructure to provide the highest degree of independence. The more regions that are utilized by a customer, the higher the level of fault tolerance may exist. Current networks consist of heavily distributed infrastructures or mainly centralized infrastructures, both of which can cause complete control plane failures when even a subset of a data zone or region loses network connectivity. The placement of control plane and data plane resources provides for optimal availability and durability during infrastructure failures.

Placement of control and data plane resources

A distributed system may implement logical control groups for distributed system resources. A request for a particular system resource may be received at a request routing tier for a distributed system that is implemented by multiple computing systems in different respective availability zones. One or more resource control nodes of a resource control tier for the distributed system that are members of a logical control group for the particular system resource may be discovered. A logical control group for the particular system resource may include at least one resource control node in a different availability zone than the availability zone that includes the particular system resource. The request may then be sent to one of the discovered resource control nodes in the logical control group in order to service the request for the particular system resource.

Logical control groups for distributed system resources

A distributed system may implement dynamic resource creation for connecting resources. Resource creation requests may be received for resources which are accessed by other resources hosted in a distributed system. An intention for the creation request is stored to perform the creation request. At a subsequent time, a request to connect another resource to the resource of the creation request is received. An infrastructure locality for the other resource may then be identified and utilized to determine a placement location for the resource. The resource may then be created at the placement location according to the stored intention. The resource may then be connected with the other resource for access.

Dynamic resource creation to connect client resources in a distributed system

A system may implement maintaining control plane data versions for a network-based service control plane. Various control plane actions may be performed which create new versions of control plane data that may be maintained for the control plane in a database. Some of these actions may be performed by multiple actors creating new versions of the same control plane data. For a particular control plane action, a new version number may be obtained to include in a new version of control plane data, and a conditional write request may be performed to insert the new version of control plane data at the database as part of an optimistic concurrency technique in order to maintain consistency for control plane data.

Maintaining versions of control plane data for a network-based service control plane

Burst throttling methods may be used manage computing resources of a data storage service. Tokens may represent I/O operations executed by a customer of the data storage service. A global token bucket may contain a set of tokens representing the overall I/O operation capacity of the data storage service. Additionally, a work token bucket may contain a set of tokens for a given logical volume maintained by the data storage service. When I/O request are received tokens may be charged the global token bucket and the work token bucket. Furthermore, if there is sufficient capacity the data storage service may charge a reduced number of tokens to the work token bucket.

Burst throttling methods may be used manage computing resources of a data storage service. Tokens may represent I/O operations executed by a customer of the data storage service. A global token bucket may contain a set of tokens representing the overall I/O operation capacity of the data storage service. Additionally, a work token bucket may contain a set of tokens for a given logical volume maintained by the data storage service. When I/O request are received the data storage service may determine if the I/O request is a member of a sequence and removed a reduced number for tokens from the work token bucket as a result. Furthermore, if there is sufficient capacity the data storage service may charge a reduced number of tokens to the work token bucket.

Burst throttling with sequential operation detection

Burst throttling methods may be used manage computing resources of a data storage service. Tokens may represent I/O operations executed by a customer of the data storage service. A global token bucket may contain a set of tokens representing the overall I/O operation capacity of the data storage service. Additionally, a work token bucket may contain a set of tokens for a given logical volume maintained by the data storage service. When I/O requests are received the data storage service may dynamically determine an amount of tokens to be removed from the global token bucket and/or work token bucket for each received request. Furthermore, if there is sufficient capacity the data storage service may charge a reduced number of tokens to the work token bucket.

Growing community of inventors

Seattle, WA, United States of America

Marc Levy