Amazon Technologies, Inc.

Chaitanya Ashok Solapurkar

Craig Wesley Howard

Ryan F Watson

Jorge Peixoto Vasquez

Anubhav Gupta

Vineet Ghatge Hemantkumar

Mukul Gupta

Jennifer Angelica Ongko

System and methods are provided for encoding and encrypting DNS labels with content metadata and client metadata. A client computing device transmits a DNS query for the encoded domain name. A DNS server receives the encoded domain name and decrypts and decodes the domain name label. The DNS server uses the decrypted and decoded client and content metadata to make a routing decision to select a particular point of presence (PoP). The DNS server sends, to the client computing device, a DNS reply with the Internet Protocol (IP) address of the selected PoP. The client computing device requests content from the PoP identified by the provided IP address.

Content-based domain name enconding, encryption, and routing system

An edge system receives requests from user devices to retrieves files from an origin server. Instead of retrieving the files as fast as possible, the edge system throttles the retrieval of files to a rate that just exceeds the speed at which the file is played by a browser or media player. The edge system determines an appropriate retrieval rate based on the contents of the file itself. For example, a manifest file associated with the file can indicate a time it takes to play back content and a bitrate of the content. Thus, the edge server can use this information to retrieve a file from an origin server at a rate that is just fast enough to minimize playback interruption. The retrieval rate determined by the edge server therefore does not rely on how fast or slow the user device retrieves the file from the edge server.

Adaptive transfer rate for retrieving content from a server

A CDN that employs a load balancer that uses probabilistic filters to estimate sizes of requested data objects and that balances incoming request loads according to the estimated sizes is provided herein. For example, the load balancer stores probabilistic filters. Each probabilistic filter is associated with a size range. When the CDN receives a data object request, the load balancer generates a cache key and tests whether the cache key is a member of any probabilistic filter. If the cache key is a member of a probabilistic filter, then the load balancer estimates a size of the requested data object based on the probabilistic filter of which the cache key is a member. The load balancer then uses the estimated size to estimate the added load on one or more cache servers. Based on the estimated added load, the load balancer selects a cache server to handle the request.

Load estimating content delivery network

Techniques are described for enabling a content delivery network (CDN) service of a cloud provider network to place content distributions across points of presence (PoPs) within a common geographic area in a manner that makes efficient use of the resources provided by the PoPs. A Domain Name System (DNS) service obtains log data reflecting requests to access distributions at various PoPs. The DNS service uses the log data to periodically generate characterization data for the distributions (e.g., in terms of requests per second, bytes transferred per second, and cache width usage). The DNS service uses the generated characterization data to assign particular distributions to particular PoPs of each PoP group such that the distributions are distributed across the PoPs according to the characterization data. The DNS service uses the assignments to generate routing data used by the DNS service to resolve DNS queries for particular distributions to particular PoPs of PoP groups.

Dynamic content placement for content delivery networks

Systems and methods are provided for managing a data store, invalidation requests, and/or resource requests. The system and methods may use one or more data structures, such as hash data structures and/or probabilistic data structures, to achieve various levels of granularity and/or computer resource efficiencies. Further, the invalidation request data may be stored at a front-end layer separate from a data store layer, such that the invalidation requests are processed based on resource requests. In the example, the hash data structure can be used for recent invalidation requests and older request data can be stored in memory-efficient probabilistic data structures. For example, probabilistic filters, such as a bloom filter, can store resource identifiers along with an associated version for space-efficient invalidation comparisons.

Growing community of inventors

Seattle, WA, United States of America

Chaitanya Ashok Solapurkar