Method and system for assembling concurrently-generated content

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application is a Continuation-in-Part of U.S. Non-provisional Application Ser. No. 09/949,532 filed Sep. 7, 2001, which is incorporated by reference herein. This application claims the benefit of U.S. Provisional Application No. 60/231,433 filed Sep. 8, 2000, which is incorporated by reference herein. This application also claims the benefit of United States Provisional Application No. 60/269,641 filed Feb. 16, 2001, which is incorporated by reference herein.

BACKGROUND

[0002] This patent specification relates generally to information retrieval and distribution systems. More specifically, it relates to a method and system for assembling and distributing content components generated in parallel by multiple component servers.

[0003] It is common for today's enterprise networks to comprise scattered arrangements of different hardware and software systems. This is due to the ever-changing data management needs of corporate enterprises, and to continuing advances in the computing hardware and software available to meet those needs. Commonly, different entities within an enterprise (for example, different departments or work sites) have disparate software applications, groupware systems, or data maintenance architectures/procedures, such that information created or maintained by one entity is not usable by another entity.

[0004] Corporate portals, also referred to as intranet portals, have been introduced to increase the accessibility and usability of information stored across the heterogeneous systems of an enterprise network. A corporate portal, which is usually overlaid onto an existing enterprise network, is designed to extract content from disparate systems on the enterprise network and to allow easier, personalized access to that content by end users. It is to be appreciated that while the features and advantages of the implementations described infra are particularly advantageous for corporate portal environments, enhancing their speed, openness, scalability, and stability, the features and advantages of the implementations are also applicable in other environments, such as with personalized “Web portals” that serve broad user bases. By way of example and not by way of limitation, one example of a corporate portal is the Plumtree Corporate Portal available from Plumtree Software, Inc. of San Francisco, Calif., while examples of personalized Web portals are typified by the MyYahoo! service from Yahoo, Inc. of Sunnyvale, Calif. and MyExcite from At Home Corp. of Redwood City, Calif. Corporate portals are also described in commonly assigned U.S. Ser. No. 09/896,039, filed Jun. 29, 2001, which is incorporated by reference herein.

[0005]FIG. 1 shows a simplified view of an exemplary user screen <b>102</b> associated with a corporate portal system, comprising a plurality of content components <b>104</b>-<b>110</b>. A content component refers to any content that is assembled, along with other content components, into a unified body of content. In the example of FIG. 1, a company news content component <b>104</b> includes an HTML display of news that is extracted, for example, from one or more company news servers, and arranged for display to the end user. A company stock quote content component <b>106</b> comprises an HTML display of a stock quote for the company and its competition that is extracted, for example, from a stock quote server. Also shown in FIG. 1 is an email content component <b>108</b> and a customer relationship management (CRM) content component <b>110</b>. According to the end user's ID <b>112</b>, the corporate portal displays the content components <b>104</b>-<b>110</b> in a personalized arrangement (for example, news at the upper left, company stock quote in the upper right, and so on) and also selects the information within each content component based on the user's ID (for example, showing the user's personal email account only, showing sports news on top of world news, showing only the user's personal CRM information, and so on). The user screen <b>102</b> of FIG. 1 would typically appear after the user (Jane Smith) has logged into the corporate portal system by supplying a user name and password.

[0006] More generally, the content components themselves can be any information communicable via any standard network protocol such as Hypertext Transfer Protocol (HTTP), Secure Hypertext Transfer Protocol (HTTPS), File Transfer Protocol (FTP), Wireless Application Protocol (WAP), and the like. Information communicable via a network includes text information, image information, Extensible Markup Language (XML), Hypertext Markup Language (HTML), or any other type of information that can be stored in a computer file, including images, sounds, and video. Throughout this specification we refer to any information sent over a network as content. We use the term content component to refer to any content that is assembled, along with other content components, into a unified body of content.

[0007] An exemplary content component is the HTML output generated by a script that communicates with an email client application. An email client application sends and receives email. Such applications usually let users compose email, and store email addresses in an address book. This script provides an HTML interface to the email client application. This script is hosted by the computer hosting the email application. This script generates HTML displaying the user's email messages, along with HTML allowing the user to compose and send email messages. This script can communicate with the email application through the application's programming interface. In this example, the HTML generated by the script is the content component (see, for example, FIG. 1, content component <b>108</b>).

[0008] Other exemplary content components are two types of HTML generated by a program that communicates with a database application. This program can be hosted by the same computer hosting the database application. The database application stores and maintains a database of information organized into records. This program can communicate with the database application via the application's interface. This program generates HTML that allows the user to search for database records. For this case, the content component is a query box. This program also generates HTML that displays database records to the user. For this case, the content component is a view of the database records (see, for example, FIG. 1, content component <b>110</b>). Further examples of content components include, but are not limited to, resources generated by a calendar application, a workflow application, a database storing proprietary personal information, a database storing proprietary business information, a database storing secure personal information, a database storing secure business information, an e-business application, and the like.

[0009]FIG. 2 shows a system <b>200</b> for delivering personalized content according to a conventional method often referred to as server-side caching. A plurality of component servers <b>202</b>-<b>206</b> provide content components to a Web server <b>208</b>. Web server <b>208</b> receives the content components in a plurality of caches <b>210</b>-<b>214</b>. Referring to FIG. 2, weather server <b>202</b> provides content components such as weather maps and forecasts into cache <b>210</b>. Stock quotes server <b>204</b> provides content components such as stock quotes and charts into cache <b>212</b>. News server <b>206</b> provides content components such as headlines and news features into cache <b>214</b>.

[0010] Users employ user terminals <b>218</b>A and <b>218</b>B through <b>218</b>N to access Web server <b>208</b> over a network <b>220</b> such as the Internet. A user establishes personalized settings in part by selecting certain of the types of content components that are routinely provided to caches <b>210</b>-<b>214</b>. Subsequent to this personalization step, the user sends a request for personalized content to main server <b>208</b>. In response, a main process <b>216</b> within Web server <b>208</b> populates a Web page with the latest cached content components according to the personalized settings for the user, and sends the personalized Web page to a user terminal <b>218</b> for display to the user.

[0011]FIG. 3 shows a system <b>300</b> for delivering personalized content according to a conventional method often referred to as client-side retrieval. A plurality of component servers <b>302</b>-<b>306</b> host various types of content components. Referring to FIG. 3, an email server <b>302</b> hosts content components such as email messages for a group of users. A stock quotes server <b>304</b> hosts content components such as stock quotes and charts. A news server <b>306</b> hosts content components such as headlines and news features. A main process <b>316</b> within Web server <b>308</b> maintains a list of the types of content components available from component servers <b>302</b>-<b>306</b>, and advertises these types of content components to users.

[0012] Users employ user terminals <b>318</b>A and <b>318</b>B through <b>318</b>N to access Web server <b>308</b> over a network <b>320</b> such as the Internet. A user establishes personalized settings by selecting certain of the types of content components that are advertised by Web server <b>308</b>. Subsequent to this personalization step, the user sends a request for personalized content to Web server <b>308</b>. In response, main process <b>316</b> populates a Web page with links, scripts, applets, or the like, that, when executed by a browser, cause the browser to retrieve the latest content components according to the personalized settings for the user. Main process <b>316</b> sends the Web page having those links, scripts, applets, etc. to the user terminal <b>318</b>, which executes the links, scripts, applets, etc. to retrieve the personalized content components from component servers <b>302</b>-<b>306</b> for display to the user.

[0013]FIG. 4 shows a system <b>400</b> for delivering personalized content according to a prior art method. A plurality of content servers <b>402</b>-<b>408</b> host various types of content. Referring to FIG. 4, a CRM server <b>402</b> hosts content such as customer lists and customer contact information. An email server <b>404</b> hosts content such as email messages for a group of users.

[0014] A stock quotes server <b>406</b> hosts content such as stock quotes and charts. A news server <b>408</b> hosts content such as headlines and news features. A main process <b>416</b> within a Web server <b>410</b> maintains a list of the types of content available from content servers <b>402</b>-<b>408</b>, and advertises these types of content to users.

[0015] Users employ user terminals <b>418</b>A and <b>418</b>B through <b>418</b>N to access Web server <b>410</b> over a network <b>420</b> such as the Internet. A user establishes personalized settings in part by selecting certain of the types of content that are advertised by Web server <b>410</b>. Subsequent to this personalization step, the user sends a request for personalized content to Web server <b>410</b>. In response, main process <b>416</b> invokes a series of processes that execute sequentially to retrieve the latest content for the content types specified by the user's personalized settings from content servers <b>402</b>-<b>408</b>. For example, referring to FIG. 4, main process <b>416</b> invokes processes <b>422</b>-<b>428</b>.

[0016] Process <b>422</b> executes first. Process <b>422</b> employs a remote procedure call (RPC) RPC<b>1</b> and a script SCRIPTI to retrieve the CRM content specified by the user's personalized settings from CRM server <b>402</b>. After the CRM content is retrieved, process <b>424</b> executes.

[0017] Process <b>424</b> employs a remote procedure call RPC<b>2</b> and a script SCRIPT<b>2</b> to retrieve the email content specified by the user's personalized settings from email server <b>404</b>. After the email content is retrieved, process <b>426</b> executes. Process <b>426</b> employs a remote procedure call RPC<b>3</b> and a script SCRIPT<b>3</b> to retrieve the stock quotes content specified by the user's personalized settings from stock quotes server <b>406</b>. After the stock quotes content is retrieved, process <b>428</b> executes. Process <b>428</b> employs a remote procedure call RPC<b>4</b> and a script SCRIPT<b>4</b> to retrieve the news content specified by the user's personalized settings from news server <b>408</b>. Main process <b>416</b> assembles the retrieved content components to form a personalized Web page, and sends the personalized Web page to the user terminal for display to the user.

[0018] One disadvantage of the approach of FIG. 4 results from the sequential execution of the retrieval processes. The overall time for processing the user's request includes the sum of the response times of the individual requests sent to the content servers <b>402</b>-<b>408</b>. If one the retrieval processes takes an unusually long time to complete or exceeds a timeout period, the overall retrieval process is delayed by that time period. Moreover, if one of the retrieval processes hangs for some reason, no content is delivered to the user at all.

SUMMARY

[0019] In general, in one aspect, the invention features a method, apparatus, and computer-readable media for satisfying a single request from a client for a plurality of content components derived from content hosted by a plurality of distinct, separately accessible component servers. It comprises receiving a single request specifying the content components; after receiving the single request, generating a plurality of information requests for the content; sending each information request to the component server hosting the content corresponding to the information request before receiving a response to any of the information requests; forming the content components from the responses to the information requests; and transmitting the content components to the client.

[0020] Particular implementations can include one or more of the following features. The single request is a request for a personalized Web page; the forming step comprises assembling the personalized Web page from the content components; and the transmitting step comprises sending the personalized Web page to the client. Implementations can comprise instantiating a timer after the step of sending each information request and before the step of forming the personalized web page; and if no response is received from one of the component servers prior to a timeout period of the timer, performing the steps of immediately establishing the response from that component server as a null value, and carrying out the steps of forming the personalized Web page and transmitting the personalized Web page to the client without waiting for that response. The component servers generate the responses in different data formats, and implementations can comprise converting the responses to a common data format. The common data format is based on a markup language. The converting step is performed at the respective component servers. The converting step is performed at a main server, the main server also receiving the single request from the user and transmitting the personalized Web page to the client. The main server is a corporate portal server. The main server is an Internet portal server. Each of the main server and the component servers are physically separate, and the information requests and responses are transmitted according to a standard network protocol. The standard network protocol is selected from the group consisting of HTTP, HTTPS, WAP, and FTP. The component servers are each selected from the group consisting of email servers, enterprise resource planning servers, and customer relationship management servers. The information requests are transmitted according to a standard network protocol. The standard network protocol is selected from the group consisting of HTTP, HTTPS, WAP, and FTP. Implementations can comprise generating a state machine to represent the progress of each information request; and recursively processing the state machines to advance the progress of each information request.

[0021] Advantages that can be seen in particular implementations include one or more of the following. Implementations issue requests for component content in parallel. This feature provides faster execution than conventional systems that issue requests sequentially. Further, if any request is unsuccessful, the content components received by the successful requests are sent to the user. In conventional systems, the failure to receive any content component could result in the delivery to the user of no content at all. Implementations incorporate a timeout feature that limits the maximum time a user must wait for a content request to be fulfilled. If any content component has not been received by the end of the timeout period, the content components gathered up to that point are sent to the user without further delay.

[0022] Implementations feature interfaces with component servers that provide cross-platform integration even when content resides on disparate, incompatible systems (for example, CORBA, Java, Microsoft, mainframes) and standardized access to data, for example, using HTTP protocol and XML content. Implementations also provide isolation of unstable content sources and access code, and increase scalability by easily distributing processing.

[0023] A description of one or more implementations are set forth in the accompanying drawings and the description below. Other features and advantages of the invention will be apparent from the description and drawings, and from the claims.

DESCRIPTION OF DRAWINGS

[0024]FIG. 1 shows a simplified view of an exemplary user screen associated with a corporate portal system, comprising a plurality of content components.

[0025]FIG. 2 shows a system for delivering personalized content according to a prior art method.

[0026]FIG. 3 shows a system for delivering personalized content according to a prior art method.

[0027]FIG. 4 shows a system for delivering personalized content according to a prior art method

[0028]FIG. 5 shows a system for delivering personalized content according to one implementation.

[0029]FIG. 6 shows a system for delivering personalized content according to one implementation.

[0030]FIG. 7 shows a system for delivering personalized content according to one implementation.

[0031]FIG. 8 shows a system for delivering personalized content according to one implementation.

[0032]FIGS. 9 and 12-<b>14</b> depict the issuing of parallel requests according to one implementation.

[0033]FIG. 10 is a flowchart showing a process for processing the requests according to one implementation.

[0034]FIG. 11 is a flowchart showing a process for processing sockets for the requests according to one implementation.

[0035]FIG. 16 shows a process used by the main server to assemble a collection of content according to one implementation.

[0036]FIG. 16. shows a process used by the main server to formulate the requests to be issued to the component servers in accordance with one implementation.

[0037] Like reference symbols in the various drawings indicate like elements.

DETAILED DESCRIPTION

[0038]FIG. 5 shows a system <b>500</b> according to one implementation. A plurality of component servers <b>502</b>-<b>508</b> host different types of content components. Referring to FIG. 5, a CRM server <b>502</b> hosts content such as customer lists and customer contact information. An email server <b>504</b> hosts content such as email messages for a group of users. A stock quotes server <b>506</b> hosts content such as stock quotes and charts. A news server <b>508</b> hosts content such as headlines and news features. A main process <b>516</b> within a main server <b>510</b> maintains a list of the types of content available from content servers <b>502</b>-<b>508</b>, and advertises these types of content to users. Of course, other types of content, such as enterprise resource planning content, can be made available to users.

[0039] Users employ user terminals <b>518</b>A and <b>518</b>B through <b>518</b>N to access main server <b>510</b> over a network <b>520</b> such as the Internet. As used herein, “user terminal” refers to any device that a user could employ to access the main server including a computer running a Web browser, a personal digital assistant, a cellular phone, and the like.

[0040] Main server <b>510</b> communicates with each component server <b>502</b>-<b>508</b> using a standard protocol such as HTTP. In one implementation, main server <b>510</b> uses the same protocol for all of the component servers. Any needed protocol translations are performed at the component server. Referring to FIG. 5, main server <b>510</b> includes one or more HTTP client libraries <b>530</b>. Each component server <b>502</b>-<b>508</b> contains includes an HTTP host library <b>532</b>. Together libraries <b>530</b> and <b>532</b> facilitate communication between the main and component servers.

[0041] Each component server may operate under a different protocol. For this reason, each component server includes a remote procedure call (RPC) and a script. The RPC and script collect the requested content components and perform any necessary protocol and data format translations. CRM component server <b>502</b> employs a remote procedure call RPC<b>1</b> and a script SCRIPT<b>1</b> to retrieve CRM content. Email server <b>504</b> employs a remote procedure call RPC<b>2</b> and a script SCRIPT<b>2</b> to retrieve email content. Stock quotes server <b>506</b> employs a remote procedure call RPC<b>3</b> and a script SCRIPT<b>3</b> to retrieve stock quotes content. News server <b>508</b> employs a remote procedure call RPC<b>4</b> and a script SCRIPT<b>4</b> to retrieve news content. Main process <b>516</b> assembles the retrieved content components to form a personalized Web page, and sends the personalized Web page to the user.

[0042]FIG. 6 shows a system <b>600</b> according to an implementation featuring separate intermediate servers. According to this implementation, an intermediate server is provided for each component server. Each intermediate server includes a HTTP host library <b>532</b>, a remote procedure call (RPC) and a script that function as described above. Referring to FIG. 6, an intermediate server <b>602</b> employs a remote procedure call RPC<b>1</b> and a script SCRIPT<b>1</b> to retrieve CRM content from CRM server <b>502</b>. An intermediate server <b>604</b> employs a remote procedure call RPC<b>2</b> and a script SCRIPT<b>2</b> to retrieve email content from email server <b>504</b>. An intermediate server <b>606</b> employs a remote procedure call RPC<b>3</b> and a script SCRIPT<b>3</b> to retrieve stock quotes content from stock quotes server <b>506</b>. An intermediate server <b>604</b> employs a remote procedure call RPC<b>4</b> and a script SCRIPT<b>4</b> to retrieve news from news server <b>508</b>. Main process <b>516</b> assembles the retrieved content to form a personalized Web page, and sends the personalized Web page to the user.

[0043]FIG. 7 shows a configuration <b>700</b> according to another implementation. According to this implementation, the scripts execute at the main server <b>510</b>, and the RPCs execute at the component servers <b>502</b>-<b>508</b>.

[0044]FIG. 8 shows a configuration <b>800</b> according to an implementation featuring split scripts. According to this implementation, each script is split into two scripts, with one script executing at the component server and the other script executing at the main server.

[0045] Referring to FIG. 8, script SCRIPTI located at main server <b>510</b> operates together with script SCRIPT<b>1</b>′ and remote procedure call RPC<b>1</b> located at CRM server <b>502</b> to retrieve CRM content. Script SCRIPT<b>2</b> located at main server <b>510</b> operates together with script SCRIPT<b>2</b>′ and remote procedure call RPC<b>2</b> located at email server <b>504</b> to retrieve email content. Script SCRIPT<b>3</b> located at main server <b>510</b> operates together with script SCRIPT<b>3</b>′ and remote procedure call RPC<b>3</b> located at stock quotes server <b>506</b> to retrieve stock quotes content. Script SCRIPT<b>4</b> located at main server <b>510</b> operates together with script SCRIPT<b>4</b>′ and remote procedure call RPC<b>4</b> located at news server <b>508</b> to retrieve news content. Main process <b>516</b> assembles the retrieved content components to form a personalized Web page, and sends the personalized Web page to the user.

[0046] Each user can request a personalized set of content components from main server <b>510</b>, including at least some content that is specific to the user (such as e-mail). Main server <b>510</b> then issues information requests for these components to the appropriate component servers <b>512</b>-<b>518</b>, which concurrently generate the requested content components. Immediately after a component has been generated by its component server, the component is sent via a standard network protocol to main server <b>510</b>. After either all of the components have been generated and communicated, or a specified timeout period has elapsed, main server <b>510</b> assembles the generated components into a unified body of content, and serves this content to the client system from which the original request was issued.

[0047] The Main Server

[0048] Throughout this description we refer to a single computer as the “main server.” It should be noted that the word “server” typically refers to a computer responsible for serving requests from user terminals, and little else. However, main server <b>510</b> also functions as a client to other servers; in this description these other servers are referred to as “content component servers” or simply “component servers.” These servers, in turn, may function as clients to yet other servers.

[0049] The characteristic that distinguishes the main server from any other servers that are involved is that the main server is the entry point to the entire system. In some implementations multiple main servers are used to meet the needs of a large number of users. In this case, all of the main servers used are similar. Load balancing software or hardware is used to distribute client requests among the available main servers. Also, substantially all of these main servers share a database of information. In this way, the state of the system is indistinguishable to users, regardless of which main server they are interacting with on any particular occasion.

[0050] An HTTP Implementation

[0051] One implementation uses the HTTP network protocol to communicate requests for content from user terminals to the main server, and from the main server to the component servers. One implementation also uses the HTTP protocol to communicate content from the component servers to the main server, and from the main server to the user terminals. HTTP offers the advantage of being the most widely used protocol. The HTTPS network protocol could also be used to implement a more secure system. The HTTPS protocol encrypts information during transmission and therefore offers greater security than the HTTP protocol.

[0052] In one implementation, the main server communicates with one or more user terminals and a plurality of component servers over TCP/IP connections established over a network. This system is suitable for implementing HTTP-based network services. The HTTP protocol is described in detail in “Hypertext Transfer Protocol—HTTP/1.0,” Network Working Group, May 1996.

[0053] Network Setup

[0054] Now system <b>500</b> is discussed in greater detail. While this discussion is directed to system <b>500</b>, it also applies to other implementations, as will be apparent to one skilled in the relevant art after reading this description. Referring again to FIG. 5, multiple user terminals <b>518</b> make requests of a single main server <b>510</b>. These requests can be issued at any time. Main server <b>510</b> makes requests of multiple component servers <b>502</b>-<b>508</b>. It is the responsibility of main server <b>510</b> to determine when to make a request, to which component server a request must be made, and the exact form of the request.

[0055] Client computers <b>518</b> issue requests to main server <b>510</b>. Main server <b>510</b> issues requests for content components to component servers <b>502</b>-<b>508</b>. Component content is sent from the component servers to the main server. Main server <b>510</b> is responsible for assembling content components, and sending this assembled and processed content to user terminals <b>518</b>.

[0056] For example, a user on a user terminal requests an update of his personal collection of content components. This request can be made by directing a standard Web browser capable of making HTTP requests to an URL. This URL represents the location from which all users of this example system obtain assembled content. Appended to this URL is the ID of the particular user making the request, for example http://MainServer/portal.asp?UserID=<b>213</b>. Standard session management techniques that are well-known to those of ordinary skill in the art are used to associate a particular user with a session on the Web server. Web development environments such as Active Server Pages (“ASP”) and Java Server Pages (“JSP”) manage session state automatically. In one implementation, all users of this system visit the same URL for updated content, but to each URL is appended a distinct user ID. In this example the ID <b>213</b> is associated with the user making the request. The main server receives a request and extracts the ID of the user that made the request. From this ID, the main server knows which user is making the request. In previous interactions with the main server, the user has specified which content components this user wishes to view. The main server is responsible for obtaining and storing this information. In one implementation, the main server provides an HTML form allowing users to select the components they wish to view from a library of components. In this example, suppose that the user that issued this request wishes to view content components A, B, and C. The main server knows that content component A can be obtained at the URL http://CS1/A.asp, content component B can be obtained at the URL http://CS2/B.asp, and content component C can be obtained at the URL http:/ /CS3/C.asp. In this case, each content component is housed on a separate component server: CS1, CS2, or CS3. But it could be the case that multiple content components are housed on the same component server. Also, these component servers could be physically located on the same local network as the main server. Component servers could also be located on a network physically separated from that of the main server. The HTTP communication protocol allows for communication between remote computers. In one implementation, the system administrator registers the content component with the main server by specifying the URL. In this implementation, the URL is stored in a relational database.

[0057] The main server then proceeds to request in parallel updated content components from these URLs. The component servers then concurrently generate their components. In some cases, the applications feeding these component servers generate HTML natively. In other cases, the component servers convert (for example, translate) the initial non-HTML content into HTML content. The component servers then post the content of these components back to the main server. The main server then receives these components, and assembles them into a unified body of content. If a component received by the main server complies with the HTML format, then the main server simply splices this component's content into a table element within a complete HTML page. Within this table, other content components are spliced into other table elements. If a received component complies with the XML format, then the main server applies an XSL style sheet to transform the XML into HTML which could then be treated just like an HTML component, and spliced into a table element. Once assembled, this table is then posted back to the user terminal from which the original request was issued. Note that the response (for example, table) is not limited to the HTML format; the response could also present data in any other mark-up or display language including, but not limited to, WML, HDML, or VoiceXML.

[0058] Parallel Requests

[0059] FIGS. <b>9</b>-<b>12</b> depict the issuing of parallel requests according to one implementation. In this implementation, A main server <b>902</b> issues requests in parallel, and waits either for the arrival of responses from all of the component servers <b>904</b>, or for the timeout period to expire.

[0060] In the first step of this example, as shown in FIG. 9, main server <b>904</b> issues four requests to four component servers <b>904</b>A, <b>904</b>B, <b>904</b>C, and <b>904</b>D. In one implementation, the issuing of requests is implemented as follows: the main thread of execution spawns four worker threads, one worker thread for each request. Each worker thread executes a process that obtains both the length of the timeout period for its particular request, and the specific request to be made. Each worker thread then issues its request. In another implementation, a single process obtains both the specific request to be made and the length of the timeout period for each request. The process then issues the requests in a rapid sequence.

[0061] In one implementation, the worker threads or processes each use a standard HTTP client library to issue requests. Some possible HTTP client libraries that could be used for this step are WinInet, libwww, or JDK. These libraries offer similar functionality. Each of these libraries offers functions that take a URL as an argument and return content downloaded from that URL. The particular client library which would most likely be used for a given implementation of this invention depends on the platform on which the main server is implemented. For example, if the main server operates on Windows NT, then the WinInet library would most likely be used.

[0062] It may be advantageous to create a customized HTTP client library that makes more efficient use of the host system's available resources. The standard libraries listed above are optimized for communications involving a single-user client application, rather than a multi-user application functioning as both a client and a server. A customized HTTP library could create several efficiencies. It might reduce the number of worker threads the main server requires to maintain HTTP connections. It might reduce the number of times network connections need to be opened and closed. It might increase speed by optimizing network address lookups.

[0063] It should be noted that an HTTP client library designed for multi-user, multi-server environments would have benefits beyond the aggregation of content on a personalized Web page. It would be useful in any situation where parallel processing of HTTP requests is desirable. Examples might include, but would not be limited to, issuing query requests to multiple query engines, aggregating feeds from XML-generating applications, or batch-posting data to a large number of Web-based forms simultaneously.

[0064] The following is a description of one implementation of such an HTTP client library. In this description, the term “user” refers to a programmer using the client library to write software programs.

[0065] The HTTP client library defines following basic objects:

[0066] HTTPRequest—This is the only user-level (that is, normally accessed by the user of the HTTP client library) object in the library. It encapsulates basic HTTP protocol methods/properties (such as header/body creation, sending actual request to the server, decoding server response, and the like). Unlike existing requests in existing HTTP client libraries, it has the capability to be linked with other HTTPRequest objects in a chain, which can be processed from the user perspective as a single HTTPRequest (work is done on all requests in parallel, from the user's perspective).

[0067] AddrResolver—Internal object, responsible for resolving URL into corresponding InetHost objects. AddrResolver maintains a cache of InetHost objects (allocates duplicate objects if necessary; frees those which are not being used). It also handles Web Proxies.

[0068] InetHost—Internal object, encapsulating a Web server. Responsible for establishing/terminating TCP/IP connections to the server, handling SSL, and tunneling Web proxies.

[0069] Here is how the objects are typically used:

[0070] 1. Create a new HTTPRequest object, specifying HTTP method and target Web server.

[0071] 2. Add necessary HTTP headers to the request.

[0072] 3. Add HTTP body, if necessary.

[0073] 4. (optional, repeat as needed) Repeat steps 1-3, link new request to the request previously created.

[0074] At this point we have a chain of the requests, containing one or more HTTPRequest objects.

[0075] 5. Invoke Process( ) method on the first HTTPRequest object, specifying desired timeout value. The method returns if either of following conditions are requests in the chain are met: a) All requests in the chain are finished. b) Timeout expires.

[0076] Throughout this process, an HTTPRequest object is in one of five request states:

[0077] 1. NEW

[0078] 2. WAITING FOR HOST

[0079] 3. SENDING TO HOST

[0080] 4. RECEIVING DATA

[0081] 5. FINISHED

[0082] The following pseudo-code details the Process( ) method: <tables id="TABLE-US-00001" num="1"><table frame="none" colsep="0" rowsep="0" pgwide="1"><tgroup align="left" colsep="0" rowsep="0" cols="1"><colspec colname="1" colwidth="273PT" align="left" /><thead><row><entry /></row><row><entry namest="1" nameend="1" align="center" rowsep="1" /></row></thead><tbody valign="top"><row><entry>while (not all requests are finished and timeout not expired) do</entry></row><row><entry>{</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="14PT" align="left" /><colspec colname="1" colwidth="259PT" align="left" /><tbody valign="top"><row><entry /><entry>for each request in a chain</entry></row><row><entry /><entry>{</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="28PT" align="left" /><colspec colname="1" colwidth="245PT" align="left" /><tbody valign="top"><row><entry /><entry>switch (state of the request)</entry></row><row><entry /><entry>{</entry></row><row><entry /><entry>case “New”:</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="42PT" align="left" /><colspec colname="1" colwidth="231PT" align="left" /><tbody valign="top"><row><entry /><entry>change state to “Waiting for host” (makes request uneditable)</entry></row><row><entry /><entry>break</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="28PT" align="left" /><colspec colname="1" colwidth="245PT" align="left" /><tbody valign="top"><row><entry /><entry>case “Waiting for host”:</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="42PT" align="left" /><colspec colname="1" colwidth="231PT" align="left" /><tbody valign="top"><row><entry /><entry>If (the host, corresponding to the request's UIRL is not immediately available)</entry></row><row><entry /><entry>{</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="56PT" align="left" /><colspec colname="1" colwidth="217PT" align="left" /><tbody valign="top"><row><entry /><entry>Try to obtain a connection (socket) to the local address resolution server.</entry></row><row><entry /><entry>If successful, send an address resolution request to it.</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="42PT" align="left" /><colspec colname="1" colwidth="231PT" align="left" /><tbody valign="top"><row><entry /><entry>}</entry></row><row><entry /><entry>else</entry></row><row><entry /><entry>{</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="56PT" align="left" /><colspec colname="1" colwidth="217PT" align="left" /><tbody valign="top"><row><entry /><entry>try to obtain a connection (socket) to corresponding web server (possibly</entry></row><row><entry /><entry>through proxy)</entry></row><row><entry /><entry>change state to “Sending to host”</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="42PT" align="left" /><colspec colname="1" colwidth="231PT" align="left" /><tbody valign="top"><row><entry /><entry>}</entry></row><row><entry /><entry>break</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="28PT" align="left" /><colspec colname="1" colwidth="245PT" align="left" /><tbody valign="top"><row><entry /><entry>case “Finished”:</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="42PT" align="left" /><colspec colname="1" colwidth="231PT" align="left" /><tbody valign="top"><row><entry /><entry>Remove request from the chain</entry></row><row><entry /><entry>break</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="28PT" align="left" /><colspec colname="1" colwidth="245PT" align="left" /><tbody valign="top"><row><entry /><entry>}</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="14PT" align="left" /><colspec colname="1" colwidth="259PT" align="left" /><tbody valign="top"><row><entry /><entry>}</entry></row><row><entry /><entry>if (any connection (socket) is ready to be used) (this check is done simultaneously on all</entry></row><row><entry /><entry>connections)</entry></row><row><entry /><entry>{</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="28PT" align="left" /><colspec colname="1" colwidth="245PT" align="left" /><tbody valign="top"><row><entry /><entry>for each request, corresponding to “ready to be used”connection</entry></row><row><entry /><entry>{</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="42PT" align="left" /><colspec colname="1" colwidth="231PT" align="left" /><tbody valign="top"><row><entry /><entry>switch (state of the request)</entry></row><row><entry /><entry>{</entry></row><row><entry /><entry>case “Waiting for host”:</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="56PT" align="left" /><colspec colname="1" colwidth="217PT" align="left" /><tbody valign="top"><row><entry /><entry>release connection to the local address resolution server</entry></row><row><entry /><entry>break</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="42PT" align="left" /><colspec colname="1" colwidth="231PT" align="left" /><tbody valign="top"><row><entry /><entry>case “Sending to host”:</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="56PT" align="left" /><colspec colname="1" colwidth="217PT" align="left" /><tbody valign="top"><row><entry /><entry>send HTTP data to the component server</entry></row><row><entry /><entry>change request state to “Receiving data”</entry></row><row><entry /><entry>break</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="42PT" align="left" /><colspec colname="1" colwidth="231PT" align="left" /><tbody valign="top"><row><entry /><entry>case “Receiving data”:</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="56PT" align="left" /><colspec colname="1" colwidth="217PT" align="left" /><tbody valign="top"><row><entry /><entry>receive response data from the component server</entry></row><row><entry /><entry>release connection to the component server</entry></row><row><entry /><entry>change request state to “Finished”</entry></row><row><entry /><entry>break</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="42PT" align="left" /><colspec colname="1" colwidth="231PT" align="left" /><tbody valign="top"><row><entry /><entry>}</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="28PT" align="left" /><colspec colname="1" colwidth="245PT" align="left" /><tbody valign="top"><row><entry /><entry>}</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="OFFSET" colwidth="14PT" align="left" /><colspec colname="1" colwidth="259PT" align="left" /><tbody valign="top"><row><entry /><entry>}</entry></row></tbody></tgroup><tgroup align="left" colsep="0" rowsep="0" cols="1"><colspec colname="1" colwidth="273PT" align="left" /><tbody valign="top"><row><entry>}</entry></row><row><entry namest="1" nameend="1" align="center" rowsep="1" /></row></tbody></tgroup></table></tables>

[0083] In one implementation, a technique referred to herein as “connection pooling” is employed. According to this technique, a socket used for an HTTP request is not automatically discarded after the response to the HTTP request is received. Instead, the socket is cached in a “connection pool” for a period of time. The Process( ) method attempts to use these sockets for subsequent HTTP requests by using the host name and port number of the HTTP request as the key to the cache. Some implementations use a separate connection pool for secure HTTP requests.

[0084]FIG. 10 is a flowchart showing the Process( ) method according to one implementation <b>1000</b>. At the point at which the Process( ) method is invoked, a chain of HTTPRequest objects has been created. In one implementation, the chain is implemented as a linked list of the HTTPRequest objects. Associated with each object in the chain is a single component server responsible for serving the request. In some implementations, the HTTPRequest objects in the chain represent the information components required to satisfy a request for a personalized Web page. In other implementations, the HTTPRequest objects in the chain represent information components required to satisfy other collections of requests for content components.

[0085] Process <b>1000</b> first determines whether there are any HTTPRequest objects in the chain (step <b>1002</b>). If not, then process <b>1000</b> releases all of the sockets (step <b>1004</b>), and finishes. If any HTTPRequest objects remain in the chain, process <b>1000</b> determines whether a predetermined timeout period has expired (step <b>1006</b>). If the timeout period has expired, then process <b>1000</b> releases all of the sockets by returning the sockets to the connection pool (step <b>1004</b>), and finishes. If the timeout period has not expired, then process <b>1000</b> examines the first HTTPRequest object in the chain (step <b>1008</b>). In one implementation, each HTTPRequest object is a state machine having the multiple states described above.

[0086] If the state of the HTTPRequest object is NEW (step <b>1010</b>), then the process generates an HTTP request by completing the HTTP headers of the HTTP request (step <b>1012</b>), and changes the state of the HTTPRequest object to WAITING FOR HOST (step <b>1014</b>). An http request is a set of headers that specify parameters of the message. Some of the parameters are set by the HTTPRequest object, such as message length, while others are set by the main server, such as what is sent, where it is sent, form of the reply, and the like.

[0087] If the state of the HTTPRequest object is WAITING FOR HOST (step <b>1016</b>), process <b>1000</b> determines whether the name of the component server has been resolved (step <b>1018</b>). If the name of the component server has not been resolved, process <b>1000</b> resolves the name of the component server (step <b>1020</b>). In one implementation, process <b>1000</b> resolves the name of the component server by contacting a local name resolution server. In some implementations, sockets to the local name resolution are cached in the connection pool in the same way that sockets to the component servers are cached.

[0088] Process <b>1000</b> then determines whether any HTTPRequest objects remain in the chain (step <b>1032</b>). If any HTTPRequest objects remain in the chain, then process <b>1000</b> examines the next HTTPRequest object in the chain (step <b>1034</b>). If no HTTPRequest objects remain in the chain, then process <b>1000</b> processes the sockets for the HTTPRequest objects (step <b>1036</b>).

[0089] However, if at step <b>1018</b>, the name of the component server has been resolved, process <b>1000</b> attempts to obtain a socket connecting to the component server (step <b>1022</b>). Process <b>1000</b> attempts to obtain a socket to the component server from the connection pool. If the connection pool contains no socket to the component server, process <b>1000</b> attempts to create a new socket to the component server. Process <b>1000</b> then determines whether a socket was obtained (step <b>1024</b>). If a socket was obtained, process <b>1000</b> changes the state of the HTTPRequest object to SENDING TO HOST (step <b>1026</b>) and adds the socket to a ready list. The ready list includes sockets that are ready to be written with outgoing data, such as the socket added to the list in step <b>1026</b>, and sockets with incoming data ready to be read, as is well-known in the relevant arts. Process <b>1000</b> then resumes at step <b>1032</b>.

[0090] If the state of the HTTPRequest object is FINISHED (step <b>1028</b>), process <b>1000</b> removes the HTTPRequest object from the chain (step <b>1030</b>). Process <b>1000</b> then resumes at step <b>1032</b>. If the state of the HTTPRequest object is not WAITING FOR HOST, SENDING TO HOST or FINISHED, process <b>1000</b> resumes at step <b>1032</b>.

[0091]FIG. 11 is a flowchart showing a process <b>1100</b> for processing sockets according to one implementation. In one implementation, process <b>1100</b> is employed during step <b>1036</b> of process <b>1000</b>. Process <b>1100</b> obtains a set of ready sockets (step <b>1102</b>). In one implementation, the set of ready sockets includes the sockets added during step <b>1026</b> of process <b>1000</b>. Process <b>1100</b> determines whether there are any sockets in the set (step <b>1104</b>). If not, process <b>1100</b> finishes. But if there are any sockets in the set, process <b>1100</b> selects one of the sockets (step <b>1106</b>), and determines whether there is a HTTPRequest object that corresponds to the selected socket (step <b>1108</b>). If not, the selected socket is removed from the set (step <b>1110</b>), and process <b>1100</b> resumes at step <b>1104</b>. But if there is a HTTPRequest object that corresponds to the selected socket, process <b>1100</b> examines that HTTPRequest object (step <b>1112</b>).

[0092] If the state of the HTTPRequest object is WAITING FOR HOST (step <b>1114</b>), process <b>1100</b> releases the socket to the local address resolution server by returning the socket to the connection pool (step <b>1116</b>). The socket to the component server is then removed from the set of ready sockets (step <b>1110</b>), and process <b>1100</b> resumes at step <b>1104</b>.

[0093] If the state of the HTTPRequest object is SENDING TO HOST (step <b>1118</b>), then process <b>1100</b> sends the HTTP request corresponding to the HTTPRequest object to the component server identified by the request (step <b>1120</b>), and changes the state of the HTTPRequest object to RECEIVING DATA (step <b>1122</b>). The socket is then removed from the set (step <b>1110</b>), and process <b>1100</b> resumes at step <b>1104</b>.

[0094] If the state of the HTTPRequest object is neither WATING FOR HOST nor SENDING TO HOST, then the state of the HTTPRequest object is RECEIVING DATA. In this case, process <b>1100</b> determines whether all of the data has been received for the HTTPRequest object (step <b>1124</b>). If not, the socket is removed from the set (step <b>1110</b>), and process <b>1100</b> resumes at step <b>1104</b>. But if all of the data has been received for the HTTPRequest object, process <b>1100</b> releases the socket by returning the socket to the connection pool (step <b>1126</b>), and changes the state of the request to FINISHED (step <b>1128</b>). The socket is then removed from the set (step <b>1110</b>), and process <b>1100</b> resumes at step <b>1104</b>.

[0095] Further elaboration on the use of the term “parallel” is needed. It should be noted that a single-processor server supporting multiple threads of execution devotes some amount of processing time to a particular thread before performing a context switch, during which computing resources are handed over to another thread. At some point during the execution of each of the worker threads described above, the thread will call upon the HTTP client library to make an HTTP request of the appropriate component server. Computing resources may then be switched over to another worker thread which may then execute its HTTP request. When examined on this level of detail it may be noted that the HTTP requests are not, in fact, issued in parallel, but instead are issued sequentially but extremely rapidly. However, examining the system on a more general and functional level, reveals that the amount of time that elapses between the issuing of HTTP requests to component servers is negligible in contrast to the amount of time likely consumed by the round-trip transmission of HTTP requests to and from the component servers, added to the amount of time consumed by the generation of content by component servers. For example, assuming the worker threads are spawned at approximately the same time, and the platform hosting the main server switches context every 100 microseconds (10-6), using a round-robin scheduling algorithm which distributes computing resources evenly amongst threads, multiple HTTP requests for content components are likely to be issued within one {fraction (1/1000)}^thof a second. The time for HTTP requests to travel across the local network to component servers could be as little as {fraction (1/1000)}^thof a second, but sending HTTP requests could also take multiple seconds. The amount of time consumed by this step is largely unpredictable because of fluctuations in network conditions. It should also be noted that component servers are not necessarily located on the local network, in which case even greater variability in the amount of time needed for request transmission is introduced. Turning to the generation of content itself, the least amount of time required to generate content is roughly {fraction (1/1000)}^thof a second to generate a static HTML page, but in general, the generation of content components will require substantially more time. Again, the amount of time consumed by this step is largely unpredictable. In summary, the amount of time between the issuing of HTTP requests to component servers is of short and consistent duration, whereas the amount of time required for request transmission and component generation varies greatly and unpredictably from component to component, and may take an arbitrarily long period of time. Immediately after HTTP requests are issued the requested content components are truly generated in parallel. For this reason, we use the term “parallel” to describe the entire process of issuing requests for content components, even if a particular implementation of this system is not capable of issuing HTTP requests in parallel. A similar analysis applies to implementations that employ a single process to issue the requests in a rapid sequence.

[0096] Referring to FIG. 12, the second step is an intermediate point in the processing of the requests. At this point, all requests for components have been issued. Component server <b>904</b>C has finished processing its request, and has accordingly returned the resulting content component. Main server <b>902</b> receives this content component, stores it, and awaits for remaining content components to be returned. The remaining component servers <b>904</b> are still processing their respective requests.

[0097] Referring to FIG. 13, the third step is the final point in the processing of the requests. Servers <b>904</b>B and <b>904</b>D complete their requests and return the resulting components. Component server <b>904</b>C has previously finished. Component server <b>904</b>A encounters a serious error and is unable to communicate any response at all to the main server.

[0098] Referring to FIG. 14, in the fourth step main server <b>902</b> communicating the resulting content, processed and assembled, to the user terminal from which the original request was issued. This content is assembled from components generated by component servers <b>904</b>B, <b>904</b>C, and <b>904</b>D. In this example, main server <b>902</b> needed to wait the full duration of the timeout period before assembling content because component server <b>904</b>A was unable to respond. If component server <b>904</b>A had been able to respond with an error message, then main server <b>902</b> could have proceeded to assemble and return content before the end of the timeout period.

[0099] Generating multiple content components in parallel can require much less time than generating the same components sequentially. As opposed to generating all content on a single server, offloading the generation of content components to separate component servers allows for more flexibility and stability in several ways.

[0100] Each component server can be configured to optimize the generation of its content component. This might include running intensive applications that should not run on the main server for performance reasons.

[0101] For example, a content component that provides an interface to a database application may need to run on the same computer hosting the database application. Saving and retrieving records to and from a database requires CPU processing and memory usage, and possibly disk input and output. If the database application were hosted by the same computer hosting the main server, then all of these operations would have a substantial negative impact on the performance of the main server. Offloading component generation from the main server to a specialized component server allows for the isolation of such applications. It also allows conversion of data from one format to another, which often requires substantial processing power, to take place on a separate server.

[0102] Any error encountered generating a content component only affects components generated by the same component server; other components are unaffected, and more importantly, the main server is unaffected.

[0103] Associating Users with Components and Preferences

[0104] In one implementation, the main server determines which user is making the request on the user terminal through some form of user authentication. Prompting the user to enter a usernarne and password is a common method of authenticating user identification. Other more secure methods might include retinal scanning or voice pattern analysis. Another option for user authentication is to allow the operating system running on the user terminal to perform the authentication. All multi-user computer systems have some means for determining users' identities and, as far as this invention is concerned, the means are functionally equivalent.

[0105] Identifying the user making a request allows for greater granularity in terms of security and presentation of content. Identifying the user making a request also allows individual users to store their preferences with the system. A user's preferences might include a list of that user's desired content components, as well as that user's display preferences for each component. In this case, the preferences of the particular user accessing the system play a role in determining which component servers are issued requests by the main server. In this case, the preferences of a particular user might also determine additional information that is sent to a component server along with the request for a component, further specifying to the component server how to generate a component.

[0106] For example, user identification may be communicated from the main server to a component server generating a particular component, allowing for the generation of personalized or secured content. In the example presented above, in which a content component displays a user's email messages, the content component would need to have the identity of the user making the request. This implementation allows each user of the system to see a distinct set of components, with each component appearing in accordance with each user's preferences, without requiring users to specify this information along with each request.

[0107] The Process of Collecting Content

[0108]FIG. 15 shows a process <b>1500</b> used by the main server to assemble a collection of content according to one implementation. This process is executed whenever the main server needs to ensure that every component within a set of components requested by a user is up to date.

[0109] One implementation employs a data caching strategy that prevents the main server from needing to execute process <b>1500</b> every time a client makes a request. Such a strategy can greatly reduce the amount of time needed to fulfill a user's request. An effective data caching strategy is tailored to each content component because it is likely that different components will need to be cached differently. An effective caching strategy also examines the user's preferences for each cached component. For example, a user requests component A with preference A<b>1</b>. The main Web server fulfills this request by issuing a request to the appropriate component server. The component server processes this request and returns the resulting content component to the main server. The main server then caches the content returned by the component server, indexed by the preference A<b>1</b> that was included in the request sent to obtain the content, and proceeds to return the content to the user that initiated the request. If, at a later time, a user requests component A with preference A<b>1</b>, then the main server can quickly return the previously cached content, without needing to contact the component server that previously generated that content. If a user requests component A with preference B<b>1</b>, however, then the main server needs to contact the appropriate component server because it is possible that the content generated using preference B<b>1</b> would be different than the previously cached content. One implementation also allows an administrator of the system to specify the length of time a particular component's content is stored in the cache. This is useful because it may be appropriate to store different components for different lengths of time. Components that change frequently should not be stored in the cache for long. For example, a component that opens a frequently changing database, extracts information, and displays this information should not be cached for long because it is likely that components returned from the cache do not accurately reflect corresponding components that would be generated by the component server. Components that don't change at all should be cached for as long as possible. For example, a component that displays a link to a useful Web page should be cached for as long as possible. It may also be the case that some components should not be cached at all. One implementation allows an administrator to turn caching on and off for a particular component.

[0110] Returning to FIG. 15, process <b>1500</b> formulates the requests to be issued to the component servers (step <b>1502</b>). Process <b>1500</b> determines which component servers need to be made requests of, and the forms of the requests that need to be made. For example, each user may be able to request an arbitrary set of content components for inclusion in a page. Process <b>1500</b> determines which components the user has chosen. In one implementation, users' choices of components might be stored in a relational database.

[0111] In one implementation, step <b>1502</b> includes the process <b>1600</b> described in FIG. 16. Process <b>1600</b> determines the identity of the user issuing the request (step <b>1602</b>). This information can be obtained and validated through a login/password request, or any other form of user authentication, as described above.

[0112] Process <b>1600</b> also determines which components the user wishes to view (step <b>1604</b>). This list of components can be retrieved from a database of information previously obtained from the user. This process allows for the association of every user with a set of content components. This allows users to automatically see the desired components without needing to specify these components for every session. This association also allows users to see the correct components regardless of which user terminal they use to access the system.

[0113] Process <b>1600</b> also determines the user's display preferences for each component requested (step <b>1606</b>). This information can also be retrieved from a database of information previously obtained from the user.

[0114] Process <b>1600</b> also determines which component servers are responsible for generating the requested components (step <b>1608</b>). It is possible that multiple component servers are capable of generating a single component. It is also possible that the user's preferences obtained for a component determine which component server is responsible for generating the component. For example, a single component may be capable of displaying information from one of two databases. It is the user's preference which database of information to view. One copy of the component is located on a component server that can conveniently access one of these databases, and the other copy of the component is located on a component server that can conveniently access the other database. In this case, each user's display preference for this component determines which component server is contacted by the main server.

[0115] Returning to FIG. 15, once the necessary requests have been formulated and it has been determined to which component servers these requests need to be issued, process <b>1600</b> issues the requests in parallel (step <b>1504</b>). The significant characteristic of requests being issued in parallel is that the main server does not wait for a response from one request before sending the next request. As discussed above, it may be that when examined on an arbitrarily high level of detail the requests are actually issued sequentially. It may also be that these request are in fact issued in parallel. This will vary from implementation to implementation, as different computer systems and networks have different capabilities, but the significant characteristic of this step is that the main server issues requests as quickly as possible, without waiting for responses. Issuing requests quickly, without waiting for responses allows each component server to begin generating the requested component as soon as possible, and in parallel with the other component servers generating components.

[0116] Once all of the requests have been issued, process <b>1500</b> waits for a response from any one of the component servers (step <b>1506</b>). There is preferably an arbitrary timeout period specified by a system administrator. The main server waits no longer than this period for all responses to arrive. The length of this timeout period can be set by individual users, or it can be a system-wide value. Using a timeout period prevents the main server from waiting indefinitely if any component server, for any reason, does not respond to a request.

[0117] In one implementation, process <b>1500</b> instantiates a timer after sending an information request to a component server. If no response is received from that component server prior to a timeout period of the timer, process <b>1500</b> immediately establishes the response from that component server as a null value, forms the personalized Web page and transmits the personalized Web page to the user terminal without waiting for that response.

[0118] If a response arrives while process <b>1500</b> is waiting, process <b>1500</b> saves this response and determines if responses have been collected from all of the requests issued (step <b>1508</b>). If there are still outstanding requests, process <b>1500</b> returns to step <b>1506</b> to await another response or a timeout. If all of the requests have been satisfied, then the main server proceeds to step <b>1510</b>.

[0119] Process <b>1500</b> generates error messages as needed (step <b>1510</b>). It is possible for a component server to return an error message because it could not generate the component requested. It is also possible for a component server not to respond to a request at all. It is also possible for network errors to be encountered. Other types of errors may be encountered. Process <b>1500</b> may generate an error message and display this error message in the place of the absent component.

[0120] Process <b>1500</b> assembles requested components into a unified body of content (step <b>1512</b>). In one implementation, components consist of formatted content that can be easily displayed in a variety of ways, including but not limited to a Web browser, a personal digital assistant, a cellular phone, or any other output device.

[0121] Process <b>1500</b> posts the assembled content to the user terminal that issued the original request (step <b>1514</b>). Note that this user terminal can actually be a personal digital assistant, a television set, a telephone, or any other output device. Process <b>1500</b> is done (step <b>1516</b>).

[0122] The invention can be implemented in digital electronic circuitry, or in computer hardware, firmware, software, or in combinations of them. Apparatus of the invention can be implemented in a computer program product tangibly embodied in a machine-readable storage device for execution by a programmable processor; and method steps of the invention can be performed by a programmable processor executing a program of computer code including instructions to perform functions of the invention by operating on input data and generating output. The invention can be implemented advantageously in one or more computer programs that are executable on a programmable system including at least one programmable processor coupled to receive data and instructions from, and to transmit data and instructions to, a data storage system, at least one input device, and at least one output device. Each computer program can be implemented in a high-level procedural or object-oriented programming language, or in assembly or machine language if desired; and in any case, the language can be a compiled or interpreted language. Suitable processors include, by way of example, both general and special purpose microprocessors. Generally, a processor will receive instructions and data from a read-only memory and/or a random access memory. Generally, a computer will include one or more mass storage devices for storing data files; such devices include magnetic disks, such as internal hard disks and removable disks; magneto-optical disks; and optical disks. Storage devices suitable for tangibly embodying computer program instructions and data include all forms of non-volatile memory, including by way of example semiconductor memory devices, such as EPROM, EEPROM, and flash memory devices; magnetic disks such as internal hard disks and removable disks; magneto-optical disks; and CD-ROM disks. Any of the foregoing can be supplemented by, or incorporated in, ASICs (application-specific integrated circuits).

[0123] A number of implementations of the invention have been described. Nevertheless, it will be understood that various modifications may be made without departing from the spirit and scope of the invention. For example, although an object-oriented software implementation is discussed, other software implementations may be used, as will be apparent to one skilled in the relevant art after reading this description. Accordingly, other implementations are within the scope of the following claims.