Content preloading using predictive models

• US 10,261,938 B1
• Filed: 08/31/2012
• Issued: 04/16/2019
• Est. Priority Date: 08/31/2012
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Jenkins, Jonathan A., Dye, Nathan Alan, Marshall, Bradley E., Taylor, Brett R.
• Primary Examiner(s): Rubin, Blake J
• Application Number:
  US13/601,563
• Time in Patent Office: 2,419 Days
• Field of Search: 709213
• CPC Class Codes: G06F 15/167 using a common memory, e.g. mailbox
• CPC Class Codes: G06F 16/9574 of access to content, e.g. by caching

Abstract

Methods and apparatus for content preloading using predictive models are disclosed. A system includes computing devices that generate a model to determine a predicted set of additional network content assets anticipated to be requested by a client after the client has requested a given set of assets. The devices determine, based on characteristics of a client'"'"'s content request, whether to initiate an asset preload operation using the model. If a decision is made to initiate an asset preload, the devices identify, using the model, the predicted set of assets corresponding to the request, and initiate delivery of at least a portion of the predicted set to one or more destinations on behalf of the client.

Claims

1. A system, comprising:

one or more computing devices comprising one or more respective hardware processors and memory and configured to;

derive data from a history of received network content asset requests, wherein the history of received network content asset requests includes history from a plurality of distinct clients;

generate, based at least in part on the derived data from the plurality of distinct clients, a model to predict a set of one or more additional network content assets anticipated to be requested by a client after the client has requested a given set of one or more network content assets;

prior to application of the model to predict a particular set of one or more additional network content assets to be preloaded to one or more caches in anticipation of a request by a particular client,

determine, based at least in part on a particular preload indicator asset comprising a script component of a particular set of network content assets requested by the particular client, whether to initiate an asset preload operation using the model, wherein the asset preload operation delivers one or more additional content assets to the one or more caches;

for a determination to initiate the asset preload operation,

predict, based at least in part on application of the model that was generated, the particular set of one or more additional network content assets to be preloaded to the one or more caches via the determined asset preload operation; and

initiate a delivery of at least a portion of the predicted particular set to the one or more caches on behalf of the particular client.

2. The system as recited in claim 1, wherein the one or more computing devices are further configured to:

modify the model based at least in part on an analysis of additional network content asset requests;

predict, using the modified model and based at least in part on the same requested particular set of network content assets used to predict the particular set of one or more additional network content assets, a different set of one or more additional network content assets; and

initiate a delivery of at least a portion of the predicted different set to the one or more caches on behalf of the particular client.

3. The system as recited in claim 1, wherein to generate the model, the one or more computing devices are configured to utilize contents of one or more HyperText Transfer Protocol (HTTP) header fields of the one or more received network content asset requests.

4. The system as recited in claim 1, wherein the model comprises one or more modifiable parameters, including one or more of:

(a) an input history depth parameter indicating a number of received network asset requests to be used to determine the predicted set (b) an output prediction length parameter indicating a targeted size of the predicted set (c) a probability threshold indicating a minimum predicted request probability for an asset to initiate a preload operation for the asset or (d) one or more cost constraints to be used to determine whether to initiate a preload operation.

5. The system as recited in claim 1, wherein the one or more computing devices are configured to:

determine one or more properties of one or more of (a) a browser being used by the particular client or (b) a computing device on which the browser is instantiated, including at least one performance property; and

utilize at least one property of the one or more properties to select a particular version of at least one network content asset of the predicted set to be delivered to the one or more caches.

6. The system as recited in claim 1, wherein the one or more caches comprise at least one of:

(a) a browser cache of the particular client or (b) a content cache at an edge server of a content delivery network.

7. A method, comprising:

deriving data from a history of received network content asset requests, wherein the history of received network content asset requests includes history from a plurality of distinct clients;

generating, based at least in part on the derived data from the plurality of distinct clients, a model to predict a set of one or more additional network content assets anticipated to be requested;

determining, based at least in part on a particular preload indicator asset comprising a script component of a particular set of network content assets requested by a particular client, to perform an asset preload operation that delivers one or more additional content assets to one or more destinations associated with the particular client;

subsequent to said determining to perform the asset preload operation,

predicting, based at least in part on application of the model that was generated, a particular set of additional network content assets to be preloaded to the one or more destinations via the determined asset preload operation; and

initiating delivery of at least a portion of the predicted particular set of additional network content assets to one or more destinations associated with the particular client.

8. The method as recited in claim 7, wherein the one or more destinations comprise a network browser of the particular client, further comprising:

customizing, using the model, at least a portion of scripting code to be provided for execution at the network browser in response to a request from the particular client for the script component, such that execution at the network browser of the at least a portion of scripting code results in a transmission of the at least a portion of the predicted set of assets to the network browser.

9. The method as recited in claim 7, wherein the delivery of at least a portion of the predicted set of assets comprises delivery of at least one network content asset responsive to a request for a hidden inline frame (iframe) HyperText Markup Language (HTML) element.

10. The method as recited in claim 7, wherein the delivery of at least a portion of the predicted set of assets comprises delivery of at least one network content asset responsive to a request for an HTML element embedded within a network content asset to trigger the preload operation.

11. The method as recited in claim 7, wherein the delivery of at least a portion of the predicted set of assets comprises delivery of at least one network content asset as part of a delivery server-initiated push operation, without a corresponding browser-initiated request for the at least one network content asset.

12. The method as recited in claim 7, wherein said generating the model comprises using at least one of:

(a) contents of an HTTP Referer header fields of a client network content asset request;

(b) contents of a HTTP User-Agent field of a client network content asset request;

or (c) a client Internet Protocol (IP) address.

13. The method as recited in claim 7, wherein the model comprises one or more modifiable parameters, including one or more of:

(a) an input history depth parameter indicating a number of received network content asset requests to be used to determine the predicted set (b) an output prediction length parameter indicating a targeted size of the predicted set (c) a probability threshold indicating a minimum predicted request probability for an asset to initiate a preload operation for the asset or (d) one or more cost constraints to be used to determine whether to initiate a preload operation.

14. The method as recited in claim 13, further comprising:

adjusting one or more of the modifiable parameters based at least in part on one or more metrics associated with asset preload operations.

15. The method as recited in claim 7, further comprising:

determining one or more properties of one or more of (a) a network browser being used by the particular client or (b) a computing device on which the network browser is instantiated; and

utilizing at least one property of the one or more properties to select a particular version of at least one network content asset of the predicted set to be delivered to the one or more destinations.

16. The method as recited in claim 7, wherein the at least a portion of a predicted set of assets includes at least one of:

an image, a static HTML component, and a dynamic HTML component.

17. A non-transitory, computer-accessible storage medium storing program instructions that when executed on one or more processors:

derive data from a history of received network content asset requests, wherein the history of received network content asset requests includes history from a plurality of distinct clients;

generate, based at least in part on the derived data from the plurality of distinct clients, a model to predict a set of one or more additional network content assets anticipated to be requested by a client;

determine, based at least in part on a particular preload indicator asset comprising a script component of a particular set of network content assets requested by a particular client, whether to perform an asset preload operation that delivers one or more additional content assets to one or more destinations associated with the particular client;

subsequent to a determination to perform the asset preload operation,

predict, based at least in part on application of the model that was generated, a particular set of one or more additional network content assets to be preloaded to the one or more destinations via the determined asset preload operation; and

initiate delivery of at least a portion of the predicted particular set of one or more additional assets to the one or more destinations associated with the particular client.

18. The non-transitory, computer-accessible storage medium as recited in claim 17, wherein the one or more destinations comprise a network browser of the client and the particular preload indicator asset comprises a script component, wherein the instructions when executed on the one or more processors:

customize, using the model, at least a portion of scripting code to be provided for execution at the network browser in response to a request from the client for the script component, such that execution at the network browser of the at least a portion of scripting code results in a transmission of the at least a portion of the predicted set of assets to a cache of the network browser.

19. The non-transitory, computer-accessible storage medium as recited in claim 17, wherein at least one asset of the particular set of network content assets requested by the particular client is accessible via a first website, and at least one asset of the predicted set of assets determined by the model for the particular set is accessible via a different website.

20. The non-transitory, computer-accessible storage medium as recited in claim 17, wherein the delivery of the at least a portion of a predicted set of assets comprises delivery to the client from a content source comprising one or more of:

(a) a content cache at an edge server of a content delivery network (b) a content generating server configured to provide network content to be cached at one or more servers of a content delivery network or (c) a web server tier of a multi-tier application comprising an application server tier.

21. The non-transitory, computer-accessible storage medium as recited in claim 17, wherein the model determines a first predicted set anticipated to be requested by a first client after the first client has requested a given set of one or more network content assets, and a second predicted set anticipated to be requested by a second client after the second client has requested the given set of one or more network content assets.

22. The non-transitory, computer-accessible storage medium as recited in claim 17, wherein the model determines a first predicted set anticipated to be requested by a first client after the first client has requested a given set of one or more network content assets within a first time period, and a different predicted set anticipated to be requested by the first client after the first client has requested the given set of one or more network content assets within a second time period different from the first time period.

23. The non-transitory, computer-accessible storage medium as recited in claim 17, wherein the instructions when executed on the one or more processors:

analyze network content asset request records from a plurality of clients to determine the predicted set for the particular client, wherein the set of clients whose request records are to be analyzed to determine the predicted set for the particular client is determined at least in part on a customization setting of the model.

24. The non-transitory, computer-accessible storage medium as recited in claim 17, wherein the instructions when executed on the one or more processors:

determine one or more properties of one or more of (a) a network browser being used by the particular client or (b) a computing device on which the network browser is instantiated, including at least one performance property; and

select a particular version of at least one network content asset of the predicted set to be delivered to the one or more destinations based at least one part on a property of the one or more properties.

25. The non-transitory, computer-accessible storage medium as recited in claim 24, wherein the particular version of the at least one network content asset comprises a summarized version of a network content asset.

26. The non-transitory, computer-accessible storage medium as recited in claim 17, wherein at least a portion of the model is incorporated within a network browser of the particular client.

27. The non-transitory, computer-accessible storage medium as recited in claim 17, wherein the one or more destinations comprise a cache, wherein the instructions when executed on the one or more processors:

determine, using the model, an asset to be evicted from the cache.

28. The non-transitory, computer-accessible storage medium as recited in claim 27, wherein the cache is a component of one of:

(a) an edge server of a content delivery network or (b) a network browser of the particular client.