The present invention relates generally to the field of a client-server computer network system, and in particular, to a system and method of accessing a document efficiently through web caching.
Web browsing is becoming an inseparable part of our daily life. We routinely retrieve documents from the Internet through a web browser. However, document download speeds are not as fast as desired.
There are multiple factors behind low document download speeds. First, the bandwidth of the Internet infrastructure is limited. In particular, the bandwidth of some web hosts is very limited, which limits the download speed of documents from those web hosts. Second, the hypertext transfer protocol (HTTP), the data transfer standard adopted by most web server manufacturers and web browser developers, has some inherent inefficiencies. Third, certain important recommendations published in the official HTTP protocol standard for improving document download speeds have not been implemented by manufacturers or developers or both.
Nevertheless, given the current infrastructure and HTTP implementation, it is possible to significantly increase document download speed at little extra cost. A conventional approach to speeding up document download speeds is to establish a cache in the client computer. The web browser stores downloaded files, including static images and the like, in the cache so that those files do not need to be repeatedly downloaded. Well known mechanisms are used to determine when a file in the cache must be replaced. From the on-line subscriber's perspective, the caching of static images and other static content frequently viewed by the subscriber substantially reduces the average time required for the document to be rendered on the computer monitor screen, and therefore the user feels that the document can be downloaded very quickly from its host. Unfortunately, there are certain limitations to this conventional approach. For instance, the cache associated with the web browser is often too small to store a large number of documents. Further, the web browser sometimes cannot tell whether it a document in its cache is fresh, and therefore needlessly re-downloads the document.
In addition to slow document download speeds, another common experience during web browsing is that a user may not be able to access a requested document, either because it has been removed from a web host's file system or because the web host is temporarily out of service.
It would therefore be desirable to provide systems and methods that address the problems identified above, and thereby improve users' web browsing experience.
In a method and system of providing a document from a server to a client according to one embodiment, the document is provided to the client. One or more documents referenced in the provided document are identified. Priorities are assigned to the identified documents and the documents are provided to the client according to the assigned priorities. In some embodiments the priority may be assigned according to a location in the provided document of the referenced document. In some embodiments the priority may be assigned according to a page rank associated with the referenced document. In some embodiments a priority for a document transmission may be increased. In one embodiment, the client communicates that the priority should be increased and in another a determination is made that the priority should be increased because the client is requesting the document currently being transmitted. In some embodiments, a transmission may be terminated based on a communication from the client.
In another embodiment, a system and method for serving a document includes serving the document and identifying a document referenced in the served document. A content difference may be calculated which represents a difference between two versions of the referenced document. In some embodiments the difference is between a fresh version of the document and a stale version of the document. In some embodiments the content difference is provided to the client in response to a request for the referenced document. In some embodiments the content difference is provided to the client along with a content fingerprint of an earlier version of the document.
In some embodiments, a document is served to a client and the document represents search results generated in response to search request. Search results are identified in the document and served to the client according to the search ranking.
In still another embodiment, the cache is updated by identifying documents having freshness parameters satisfying certain conditions. If the conditions have been satisfied, a new version of the documented is obtained and a content fingerprint is generated for the new version of the document. A content difference may be generated between the newly downloaded document and a previous version, and then stored.
In another embodiment, a system and method of serving a document includes receiving a request from a client for a document including a content fingerprint based on a version of the document. A response is sent back to the client that includes a content difference between that version of the document and a later version of the document. In some embodiments, the content difference is generated prior to the request.
In still another embodiment, a system and method of requesting a document includes receiving one response including the requested document and a second response including a second document and its content fingerprint. The client determines whether the content fingerprint is resident in the cache and if so, sends a communication that the second response should be terminated.
In still another embodiment, a system and method of requesting a document includes receiving a request for the document and determining whether the document is currently being received. If the document is being received, then a communication is sent to the server requesting that the document be sent with a higher priority.
For a better understanding of the nature and embodiments of the invention, reference should be made to the Description of Embodiments below, in conjunction with the following drawings in which like reference numerals refer to corresponding parts throughout the figures.
Like reference numerals refer to corresponding parts throughout the several views of the drawings.
In this embodiment, the application 104 has no associated cache or does not use its associated cache, and instead directs all user requests to the client cache assistant 106. While the following discussion assumes, for ease of explanation, that the application 104 is a web browser, the application can, in fact, be any application that uses documents whose source is a network address such as a URL (universal resource locator). Similarly, whenever the term “URL” is used in this document, that term shall be understood to mean a network address or location. In this context, the term “document” means virtually any type file that may be used by a web browser or other application, including but not limited to audio, video, or multimedia files. An advantage of the arrangement shown in
The server 120 includes at least a server cache 122 and 128. In some embodiments, the server 120 and/or the server cache 122/128 are deployed over multiple computers in order to provide fast access to a large number of cached documents. For instance, the server cache 122/128 may be deployed over N servers, with a mapping function such as the “modulo N” function being used to determine which cached documents are stored in each of the N servers. N may be an integer greater than 1, for instance an integer between 2 and 1024. For convenience of explanation, we will discuss the server 120 as though it were a single computer. The server 120, through its server cache 122/128, manages a large number of documents that have been downloaded from various hosts 134 (e.g., web servers and other hosts) over the communications network 132.
In an embodiment, the server 120 also includes an index cache 122, a DNS cache 126, an object archive 128 and a DNS master 130, which may be connected. In some embodiments, server 120 does not include the DNS cache 126 and DNS master 130. In some embodiments, these various components co-exist in a single computer, while in some other embodiments, they are distributed over multiple computers. The remote cache server 124 communicates with the other components in the server 120 as well as web hosts 134 and domain name system (DNS) servers 136 over the Internet 132. The term “web host” is used in this document to mean a host, host server or other source of documents stored at network locations associated with the web host. The remote cache server 124 may access a search engine repository 140, which caches a huge volume of documents downloaded from millions of web servers all over the world. These documents are indexed, categorized and refreshed by a search engine. The search engine repository 140 is especially helpful for satisfying a user request for a document when the connection between the remote cache server and the web host storing the document is interrupted, as well as when the web host is in operative or otherwise unable to respond to a request for the document. In some embodiments, a repository interface 138 is disposed between the remote cache server 124 and the search engine repository 140. The repository interface 138 identifies documents in the search engine repository 140 that have been determined to be stable or fresh. The repository interface 138 works with the remote cache server 124 to update the index cache 122 indicating that these documents are in the search engine repository 140.
In one embodiment, unlike the HTTP connection between a web browser and a web server, a persistent connection (sometimes herein called a dedicated connection) is established between the client cache assistant 106 and the remote cache server 124 using a suitable communication protocol (e.g., TCP/IP). This persistent connection helps to reduce the communication latency between the client cache assistant 106 and the remote cache server 124. In one embodiment, the persistent connection comprises at least one control stream and multiple data streams in each direction. A more detailed discussion of the components in the server 120 is provided below in connection with
In another embodiment, and in reference to
Referring back to
Compared with the volume of documents cached in a client 102, the volume of documents cached in the server 120 is often significantly larger, because a server often provides documents to multiple clients 102. As a result, it is impossible to store all the documents in the server's main memory. Accordingly, and referring to
Referring to
There are three possible outcomes from the client cache check (407). The result may be a cache miss, because the client cache does not have a copy of the requested document (409). A cache miss typically occurs when the user requests a document for the first time, or when a prior version of the document is no longer valid or present in the client cache (e.g., because it became stale, or the client cache became full). Otherwise, the result is a cache hit, which means that the client cache has a copy of the requested document. However, a cache hit does not guarantee that this copy can be provided to the requesting user. For example, if the timestamp of the cached copy indicates that its content might be out of date or stale, the client cache assistant may decide not to return the cached copy to the client (411). If the document content of the cached copy is deemed fresh (413), the client cache assistant identifies the requested document as well as other related documents (e.g., images, style sheet) in the client cache, assembles them together into a hypertext markup language (HTML) page and returns the HTML page back to the web browser (417). In contrast, if the cached copy is deemed stale or if there is cache miss, the client cache assistant submits a document retrieval request to a corresponding remote cache server (415).
An exemplary document retrieval request, shown in
At 504, the remote cache server identifies the IP address of the web host through a DNS lookup. Please refer to the discussion below in connection with
There are three distinct scenarios associated with the results coming out of the server object lookup (506) and the client object lookup (508) against the object archive:
In the first scenario, the server object and the client object may be identical if they share the same content fingerprint. If not, the server object is newer than the client content. The second scenario may occur when the remote cache server downloads and stores the server object in response to a previous document retrieval request from another client. Note that the freshness of the server object will nevertheless need to be evaluated before it is used to respond to the current document retrieval request. In the third scenario, the remote cache server may have never received any request for the document, or the corresponding object may have been evicted from the server's caches due to storage limitations or staleness of the object.
The server object lookup (506) comprises two phases. The first phase is to find the content fingerprint of the server object by querying the index cache using the requested document's URL fingerprint. In some embodiments, this query is quite efficient because the index cache is small enough to be stored in the server's main memory. If no entry is found in the index cache, not only is the second phase is unnecessary, there is even no need for the client object lookup, because the initial lookup results fall into the third scenario. However, if a content fingerprint is identified in the index cache, the second phase of the server object lookup is to query the object archive for the server object's content and other relevant information using the identified content fingerprint from the first phase. Meanwhile, the remote cache server may also query the object archive for the client object's content using the content fingerprint embedded in the document retrieval request, if any.
If a server object is found in the object archive (518), the remote cache server examines the server object to determine if the server object is fresh enough to use in a response to the pending document request (512). If the server object has an associated expiration date, it is quite easy to determine the freshness of the server object. If not, a secondary test may be used to determine the server object's freshness. In one embodiment, a simple test based on the document's LM-factor is used to determine the server object's freshness. The LM-factor of a document is defined as the ratio of the time elapsed since the document was cached in the object archive to the age of the document in accordance with the date/time assigned to it by its host. If the LM-factor is below a predefined threshold, e.g., 50%, the document is treated as fresh; otherwise, the document is treated as stale. However, there may also be some embodiments or situations where a document is determined to be stale according to the freshness parameters or other information and may nevertheless be used despite its age. This may occur, for instance, when a fresh copy of the document is not available from its host.
If the server object is deemed to be fresh and its content is different from that of the client object, the remote cache server generates a first content difference between the server object and the client object (514). The content difference may be generated, based on the content of the content and server objects, using any suitable methodology. A number of such methodologies are well known by those skilled in the art. Some of these methodologies are called differential compression.
If only a server object and no client object was found, the first content difference is essentially the same as the server object. At 516, the remote cache server returns the first content difference to the client cache assistant for the preparation of an appropriate response to the application. In one embodiment, the content difference is compressed by the remote cache server before being sent to the client cache assistant so as to reduce transmission time over the connection between the remote cache server and the client cache assistant. In another embodiment, compression is not used. In yet another embodiment, compression is used only predefined criteria are met, such as a criterion that a size of the content difference (or a size of the response that includes the content difference) exceeds a threshold.
When the server object is deemed not sufficiently fresh (512), or no server object is found in the object archive (518), the remote cache server retrieves a new copy of the requested document from the document's host, or in some embodiments, the search engine repository 140 (520). In the embodiments including the repository flag of table 213 described earlier, and when the repository flag is set (538), the remote cache server 124 obtains the document from the search engine repository 140 (540). In instances where the repository interface 138 and remote cache server 124 have updated the index cache 122 for a document not yet requested, the index cache 122 will contain an entry (including the repository flag to use the search engine repository 140), and yet no corresponding document copy will be resident in the object archive 128. The document is obtained from the search engine repository 140 and sent to the client cache assistant 106 (542). In some embodiments, a content fingerprint is generated for the document, the document is recorded in object archive 128, and the various tables are updated (544). Regardless of whether this document is recorded (as in 544), a new copy of the document content is obtained from the document's web host (546), a content fingerprint is generated for the document, the document is recorded in object archive 128, and the various tables are updated (548).
If the repository flag is not set or the embodiment does not include the flag, then the document is requested from the web host (521). After receiving the document, the remote cache server registers the new document in its index cache and object archive (522) as a new server object. The registration includes generating a new content fingerprint for the new document and creating a new entry in the index cache and object archive, respectively, using the new content fingerprint. A more detailed discussion of downloading documents from a web host is provided below in connection with
As mentioned above, there is no guarantee that the remote cache server will be able to download a new copy of the requested document from the web host. For example, the web host may be temporarily shut down, the web host may have deleted the requested document from its file system, or there may be network traffic congestion causing the download from the web host to be slow (e.g., the download time is projected, based on the download speed, to exceed a predefined threshold). If any of these scenarios occurs, the search engine repository 140 (
Having access to a repository copy is extremely helpful when no server/client object is identified in either the client cache or the server object archive, and access to the web host is not currently available. In this case, the repository becomes the only source for responding to the document request with a document, as opposed to responding with an error message indicating that the document is not available. Even though there is no guarantee that the repository copy always has the same content as the copy at the web host, it is still preferred to return the repository copy than to return an error message. This is especially true if the requested document has been deleted from the web host's file system. To avoid confusing the user, the client cache assistant may attach to the response a notice indicating that the document being returned may be stale.
A document download request from the remote cache server to the host of the requested document is not necessarily triggered by a user request as indicated above. In particular, the document download request may be initiated by the remote cache server independent of any request from a client computer. For instance, the remote cache server may periodically check the expiration dates of the documents cached by the remote cache server by scanning each entry in the index cache. If a document has expired or is about to expire, e.g., within a predefined expiration time window, the remote cache server will launch a download request for a new version of the document to the web host, irrespective of whether there is a current client request for the document. Such a document download transaction is sometimes referred to as “prefetching”.
Document prefetching, however, generates an entry in the web host's access log that is not tied to an actual view of the prefetched document. Therefore, in one embodiment, if a real client request for the document falls within the predefined expiration time window, the remote cache server initiates a document prefetching while responding to the user request with the “almost-expired” version of the document from the server object archive. If the prefetched version is determined to be the same as the “almost-expired” version (as determined by comparing the content fingerprints of the two document copies or versions) the remote cache server simply renews the “almost-expired” version's expiration date without taking any further action. If the prefetched version is different from the “almost-expired”version, the remote cache server generates a new content difference between the prefetched version and the “almost-expired” version and transmits this content difference to the client cache assistant. In yet another embodiment, the remote cache server not only prefetches documents from the various web hosts but also precalculates the content differences between the new server objects corresponding to the prefetched documents and the next most recent server objects in the server object archive, and caches the precalculated content differences in its object archive for later use when a user requests these documents. This feature is particularly effective when applied to those documents that are updated and visited frequently. The stored content difference could be available via the content fingerprints and indicate which contents had been compared. Prefetching is discussed in more detail referring to
In an alternative embodiment, the processes of generating the first content difference (514) and returning the first content difference (516) precede the process of determining the freshness of the server object (512). So when the remote cache server generates the second content difference (524), the client cache assistant has received or is in the process of receiving the first content difference. As a result, the second content difference is not between the new server object and the original client object, but between the new server object and the old server object (which is now the new client object). A more detailed discussion of how the remote cache server transfers multiple content differences to the client cache assistant is provided below in connection with
The first option is to search the DNS cache 126 (
If no address record is found in the DNS cache 126 (706-no), the remote cache server must query the DNS master for the IP address associated with the hostname (708). If no IP address is found (710-no), the DNS master submits a request for resolving the hostname to a DNS server 136 (714). The DNS server 136 is sometimes herein called the “third-party DNS server”, to distinguish it from the DNS master. Otherwise (710-yes), the DNS master returns the IP address found in the DNS master to the remote cache server (711) for at least submitting a document downloading request to the web server identified by the IP address.
In some embodiments, the IP address found in the DNS master may not be sufficiently fresh (712) or may soon be deemed stale. If the IP address is not stale (712-no), the DNS master may still update one or more parameters (e.g., LUT, TTL, user visit frequency) of the address record containing the IP address in the DNS master (717). For example, the DNS master may update the user visit frequency of the address record, which, as discussed below, might affect the record's self-refresh priority.
If the IP address is stale (712-yes), the DNS master will submit a request to the third-party DNS server (714). Upon receiving the IP address from the third-party DNS server, the DNS master updates itself (716), returns the IP address (718) to the requesting remote cache server and updates the corresponding address record in the DNS cache (720). While updating the appropriate address record in the DNS master, the DNS master may update one or more of the aforementioned parameters in the address record.
If the IP address found at 710-yes is deemed stale or soon to be stale, then the DNS master submits a request for resolving the hostname to the third-party DNS server 136 (714). The remote cache server may still download a document using the stale IP address returned at 711, but in this instance, it will withhold the downloaded document temporarily until a new IP address is returned by the third-party DNS server (714). If the new IP address is identical to the stale one, the downloaded document is sent to the client cache assistant. Otherwise, the remote cache server discards the temporarily withheld document and downloads another copy using the new IP address.
Note that the DNS lookups discussed above in connection with
The resolution of IP addresses for the hostnames of the embedded links in a document may trigger the DNS master to submit multiple DNS resolution requests to the third-party DNS server. It is well understood that DNS resolution through a third-party DNS server is a time-consuming process and avoided where possible. In order to improve the remote cache server's response time, the DNS master may be configured to self-refresh the address records when predefined criteria are met. In one embodiment, the DNS master prioritizes an address record's self-refresh schedule according to its TTL and user visit frequency. For instance, the DNS master may be configured to perform a self-refresh at least a few minutes before an address record's TTL expires. In some embodiments, addresses having a higher user visit frequency are be granted a higher priority for self-refresh. Other well-known prioritization strategies are contemplated. The DNS master then updates entries, if needed, in the DNS cache, thus maintaining the freshness of those entries.
In one embodiment, when a web host is found to have a new IP address that is different from the existing one stored in the DNS master, this is interpreted as indicating that the web host has also updated the content of the requested document. The remote cache server therefore terminates the other two lookups (506 and 508 in
The ultimate purpose of the DNS, server object and client object lookups (504, 506, 508) is to provide information necessary for downloading a new document from a web host. These three processes are coordinated by the remote cache server and their respective results “converge” at 520 in
An exemplary download request generated by the remote cache server is shown in
In response to the download request, the web host returns an HTTP response to the remote cache server (803) which will indicate whether the document was found and its freshness. The web host could indicate that it does not have a fresher version, the document was not found, or that it is transmitting the document because it had a fresher version (805). If the web host has a fresher version, the remote cache server proceeds to download the document as well as any associated documents from the web host (807), creates a new content fingerprint for each of the newly downloaded documents, registers the newly downloaded documents in its index cache and object archive as the new server object (522) and generates a second content difference between the new server object and the client object (or the old server object) at 524. At 526 the remote cache server returns the second content difference or the “Not Modified” response to the client cache assistant. If not (805-No), the remote cache server re-uses the existing server or client object at 813. In one particular embodiment, the remote cache server generates a content difference of null, or zero which causes the client cache assistant to use the same object when responding to the request to the application. If the server object is to be reused, the remote cache server generates a first content difference between the server object and the client object (514), and sends the first content difference to the client cache assistant (516).
In some embodiments, the remote cache server sometimes “manipulates” the content of a newly downloaded document to improve the user experience before storing the document or generating the second content difference. For instance, if the newly downloaded document comprises slow redirect instructions such as META-tag style redirects or Javascript style redirects, the remote cache server may replace them with more efficient immediate HTTP redirects. In another example, the remote cache server may eliminate certain portions of the documents deemed to be of low importance, e.g., whitespaces or comments, to make the remaining content more compressible. If the document fails to specify the parameters of an embedded image, e.g., its width or height or both, the remote cache server may modify the document to provide the missing parameter values so as to expedite the rendering of the image on the client. The remote cache server may also remove popup windows, advertisement banners or other content from newly downloaded documents either in accordance with a predefined set of rules, or in accordance with a rules specified by a particular client or user.
As mentioned above in connection in
Prior to the generation of the second content difference, the remote cache server checks the status of the transfer of the first content difference (902). If the transfer of the first content difference has advanced substantially, e.g., more than 50% of the first content difference has been received by the client cache assistant, the remote cache server may decide to let the transfer be completed. In this case, the remote cache server generates the second content difference between new server object and the old server object (906) and then transfers the second content difference to the client cache assistant (916). The generation of the second content difference overlaps, at least in part, with transmission of the first content difference to the client cache assistant.
In some embodiments, if substantial process has not yet been made with respect to the transfer of the first content difference, the remote cache server terminates the transfer (908). As a result, the client cache assistant receives an incomplete content difference. If the incomplete content difference constitutes only a minor portion of the first content difference (910), the client cache assistant may abandon it and the remote cache server, in turn, generates a new content difference between the new server object and the original client object (914). If the incomplete content difference constitutes a significant portion of the first content difference, the client cache assistant may choose to keep the incomplete first content difference to save data transmission time. Based upon the client cache assistant's decision, the remote cache server generates a content difference between the new server object and the combination of the old server object and the original client object (912).
In one embodiment, the process of loading content differences from the remote cache server to the client cache assistant is actually an on-going process independent from any specific user request for documents. For example, the remote cache server may choose to preload content differences related to those documents that are frequently requested by users through a particular client cache assistant whenever it detects a content change to a document. Since the preloading is not triggered by any particular user request, the remote cache server may assign a relatively low priority data stream to the preloading process and reserve data streams with high priorities for the loading process in response to a user request. However, the remote cache server may upgrade the priority of a data stream associated with the preloading process using a corresponding control stream once it receives a user request for a document that is currently being preloaded. Preloading is discussed in more detail in reference to
Similarly,
Finally,
Although history log 1328 and the repository interface 1330 are illustrated as part of memory 1312, these are shown within memory 1312 simply for illustrative purposes. In some embodiments, the history log 1328 and the repository interface 1330 will reside on other servers, such as servers associated with a search engine or search engine crawler.
For illustrative purposes,
Suppose that on the morning of Mar. 7, 2004, a user types a URL of a website, http://www.python.org/, into a web browser operating in a client to access the website's homepage. In this example, it is assumed that the user's last visit 1410 to the homepage occurred on Mar. 2, 2004. In other words, the client cache assistant in the client has a cached copy of the homepage that is about five days old. It is also assumed that the remote cache server's last visit 1430 to the website occurred on Mar. 5, 2004. Therefore, the remote cache server has another copy of the homepage which is two day old. For illustrative purposes, it is further assumed that the web server hosting the website has just updated the homepage 1440 a few hours ago.
In response to the user's command, the web browser submits an HTTP GET request for the homepage to the client cache assistant. In one embodiment, the GET request includes the following HTTP header fields:
Upon receipt of the GET request, the client cache assistant searches its client cache and identifies the five-day-old homepage in the cache. In this example, it is assumed that the cached homepage does not have an associated expiration date/time, which is a common phenomenon. Without an expiration date/time, and a client cache copy that is a few days old, the client cache copy is automatically assumed by the client cache assistant to be potentially stale. Thus, the client cache assistant submits a document retrieval request to the remote cache server. The document retrieval request, besides the information specified in the GET request's header, includes the following additional header fields:
The reason for including the two additional HTTP header fields, “If-Modified-Since” and “If-None-Match”, is to inform the remote cache server that it already has a five-day-old copy of the homepage and the copy's entity tag is “199faa-1842-19d9199e”. In other words, the client cache assistant is only interested in a new copy of the homepage that is fresher than what is currently in the client cache. The two fingerprints of the cached homepage are used by the remote cache server to do a server object lookup and a client object lookup.
Upon receiving the document retrieval request, the remote cache server initiates the three lookups using the hostname “www.python.org”, the content fingerprint “0x07e111eec01dda7a” and the URL fingerprint “0x0323efd056b”, respectively. For instance, the DNS lookup resolves the IP address “194.109.137.226” of the web host by querying the DNS cache, the DNS master or an external DNS server, if necessary. As mentioned above, the object archive is a mapping from a content fingerprint of a document to an actual copy of the document's content. Therefore, the client object lookup returns a copy of the homepage that is cached on the client side, which is thereafter used as the basis for generating a content difference between a server object and a client object (see, e.g., processes 514 and 524 in
The server object lookup attempts to map the URL fingerprint of the homepage to a content fingerprint in the index cache stored in the main memory. In one embodiment, the remote cache server locates an entry in the index cache including the following set of header fields:
The “Index Date” field and the “Last-Modified” field suggest that this entry was created in the index cache about one day ago, while the server object itself is about two days old. The “Expires” field is filled with “N/A” because the web host did not specify a lifetime for the server object. Since the content fingerprint of the server object “0xb5a7d7e111da7a11” is different from that of the client object “0x07e111eec01dda7a” and its content is at least three days fresher than that of the client object, the remote cache server must decide whether the two-day-old server object is sufficiently fresh. In one embodiment, the LM-factor of a server object is defined as the ratio between the age of its index entry and the age of the server object itself. The LM-factor of the server object in this example is 50% (one day vs. two days), which the remote cache server may assess as being too high (and thus too risky) for the server object to be treated as fresh (i.e., it is deemed potentially stale).
In some embodiments, the remote cache server generates the first content difference between the server object and the client object and transmits the first content difference back to the client cache assistant even if the server object is deemed potentially stale. At the same time, the remote cache server sends a download request to the web host located at the IP address of “194.109.137.226” for the homepage's current content.
When preparing the download request, the remote cache server makes the following modifications to the header fields as received from the client cache assistant:
Had the content of the homepage not been updated by the web host after March 5th, either one of the two header fields, “If-Modified-Since” or “If-None-Match, would have informed the web host that a response including the homepage's current content is unnecessary, because it will be identical to that of the server object. In this case, the web host's response is simply a HTTP response header comprising the following header fields:
In this particular example, since the homepage was updated just a couple of hours ago, it will be sent back the remote cache server to replace the two-day-old server object and the HTTP response header will be something like this:
The remote cache server deletes the index cache entry (but not the object archive entry) corresponding to the old server object, generates a new entry for the new server object in the index cache and saves the new server object in its object archive. Assuming that the transfer of the first content difference has been completed, the remote cache server accordingly generates a second content difference between the new server object and the old server object and transmits the second content difference to the client cache assistant. Finally, the client cache assistant merges the first and second content differences with the five-day-old client object to generate a new client object, which is then sent to the web browser for rendering.
In this example, the homepage includes five links to images and one link to a style sheet. As a result, it is not enough for the web browser to render the homepage using only the new client object. In one embodiment, as the new client object makes its way to the web browser, the web browser makes GET requests similar to the one for the homepage for other documents associated with the homepage. The client cache assistant checks if the associated documents can be located in its cache, and if not, sends document retrieval requests to the remote cache server, which either identifies the associated documents in its object archive or sends requests to the corresponding web hosts. In this particular example, since the client cache assistant has the five-day-old homepage in its client cache, it will likely finds a copy of each of the associated documents in its client cache. Since these associated documents are more likely to be static than the homepage itself, the client cache assistant will re-use them unless the remote cache server returns a new version of these documents.
In another embodiment, the system operates in a more preemptive fashion. In particular, the client cache assistant, includes not only the homepage's URL and content fingerprints, but also the associated documents' URLs and content fingerprints in its document retrieval request. The remote cache server, upon receipt of the document retrieval request, processes each document therein one by one, following a same set of procedures as discussed above. Consequently, the client cache assistant receives not only content differences corresponding to the homepage itself, but also content differences corresponding to the associated documents, if they have been updated as well.
The embodiments using the repository interface 138 and repository flag in table 213 are described in reference to
The repository interface 138 interfaces with the history log 1504 and the remote cache server 124. The repository interface 138 includes a stable extractor 1508. The stable extractor 1508 uses the history log 1504 to identify documents that are identified as having reached a certain level of stability such that the copy of the document in the search engine repository 140 has a high probability of being the same as the document on the document's original host. Accordingly, the remote cache server 124 should initially use the copy of the document from the search engine repository 140, which in many cases will be delivered faster. Alternatively, or in combination with the stable extractor 1508, the repository interface 138 may include a fresh extractor 1506. The fresh extractor 1506 uses the history log to identify documents in the search engine repository 140 which may have been recently retrieved and stored. For these documents it may be faster to have the documents initially delivered from the search engine repository 140 rather than from the original web host where the document originated.
Information about the fresh or stable documents identified by the repository interface 138 is then sent to the remote cache server 124 for merging into the index cache 122. This merging includes setting of the repository flag in the index cache records for the identified documents to indicate that these documents should be obtained from the search engine repository 140. This has the beneficial feature of saving the time and effort in contacting the original document host to obtain a copy of the requested document. In some embodiments, a document identified by the repository interface 138 is entered into the index cache 122 only if the index cache 122 does not already have an entry for the same document. In other embodiments the entry in the index cache 122 will be modified and the flag set when the document copy in the search engine repository 140 is identified as being fresher than the copy in the object archive 128. In some embodiments, an entry in the cache index 122 will be modified by changing one or more of the freshness parameters of the entry based on stability information obtained by the stable extractor 1508 for the corresponding document. This may occur, for example, in situations where the stability information obtained from the repository indicates a greater degree of stability than indicated by the freshness parameters in the entry. Other embodiments could combine the two approaches.
Referring to
The stable extractor 1506 works with the history log 1504 to identify those documents which have reached a level of stability by having not changed recently or not being expected to change in the near future. Accordingly, there is a high likelihood that these documents obtained from the search engine repository 140 reflect the current state of the document on the web host. For example, the stable extractor 1506 could identify those documents which have not changed in the past Y (e.g., a predefined number between 2 and 10) downloads or within a predetermined time period. In some embodiments where the server does not provide a last modified date, this date can be approximated. If the document was unchanged between two successive downloads, the earlier date can be used as an approximation of the last modified date. The earlier and later download dates can be used to compute the LM-factor of the server object (defined as the ratio of the age of its index entry (i.e., the time since the server object was last downloaded or last verified to be to be fresh) the age of the server object itself (i.e., the amount of time since the object was last modified)), which is then compared to a threshold to determine if the server object is stale. In some embodiments, the LM-factor of the object in the search engine repository may by scaled in accordance with the page rank of the object in order to preferentially create cache index entries for popular or important pages. The initially identified set of stable documents may be reduced by selecting a predefined number of those pages having the highest page ranks, or by eliminating identified documents having a page rank 1618 below a predefined threshold. Reducing the set of identified stable documents in this way may help to avoid populating the index cache with information about documents unlikely to be requested by users. The repository interface 138 sends the information about the documents to the remote cache server 124 which uses the information to populate the fields in index cache 122.
From the history log 1504, the fresh extractor 1506 can determine those documents for which the search engine repository 140 has a fresh version within it. In these instances, it may be quicker to obtain the document from the search engine repository 140 than from the document's URL host. A set of fresh documents is identified by the fresh extractor 1506 by examining one or more parameters in a set of history log records and applying a set of predefined criteria to those parameters so as to evaluate document freshness. One method of identifying fresh documents, which in some embodiments may be defined as documents downloaded within a predetermined period of time, is by examining the timestamp value 1606 in a set of history log records. In some embodiments, the number of documents within this set of identified fresh documents may be reduced by selecting a predefined number of those pages having the highest page ranks, or by eliminating identified documents having a page rank 1618 below a predefined threshold. Reducing the set of identified fresh documents in this way may help to avoid populating the index cache with information about documents unlikely to be requested by users. Alternatively, or in combination with the above, the fresh extractor 1506 could examine the expiration date/time value 1614 and determine that it will be some period of time until that document becomes stale. Documents in the identified set whose expiration is not within a predetermined time period could be considered fresh. Alternatively, or in combination, the fresh extractor 1506 could examine the date/time value 1612 and determine those documents having values within a predetermined time period from the current date which could be considered fresh. One of ordinary skill in the art could recognize many different ways to use information from the history log to identify documents which would have a high likelihood of being fresh enough to allow the copy of the document to be obtained from the search engine repository 140 instead of the original host.
Prefetching and preloading can be used in combination to decrease the latency observed by the user on client 102. For example, when the remote cache server 124 sends a document (or content difference) to the client cache assistant 106, the remote cache server 124 could examine the resources accessed by the document and send them to the client cache assistant before the resources are requested. For example, if the document sent to the client cache assistant 106 represented search results from a previously submitted query to a search engine, then prefetching might be used to obtain at the remote cache server 124 the content of the documents most likely to be selected and preloading might be used to send to the client cache assistant 106 the document contents of those search results (for example, the top three results). The remote cache server 124 would send these documents in a lower priority data stream, which would be converted into a higher priority data stream when the document is actually requested.
When a document is requested by the client cache assistant 106 or is being sent to the client cache assistant 106, the resources in it are identified (1702). If a content difference is being sent to the client assistant 106 then the most recent document content is used. The resources in the document are identified and then prioritized to determine the order and priority for obtaining and sending the resources to the client cache assistant 106 (1704). Prioritization could be done in a variety of ways based on the type of document. For example, as mentioned above, if the document represented search results, then the documents associated with the search results, or a subset thereof comprising the N most highly ranked documents in the search results, could be identified and sent in the order of presentation in the document. Another example might order the documents based on the order of the resource links in the document, e.g., those resources earlier in the document would be obtained before or with a higher priority than those later in the document. The documents would then be obtained (1706) by any of the methods previously described (e.g., examining the index cache and obtaining each of the documents from either the object archive 128, web host 134, or search engine repository 140) and any associated processing would occur (e.g., generating content fingerprints and updating tables).
For a particular document content to be sent, the remote cache server 124 would open a channel, or use an existing one, to send the document's URL fingerprint and content fingerprint in the portion of the data stream prior to the document content (1708). Initially, this stream would be assigned a low priority and, in one embodiment, this is communicated to the client cache assistant 1206 via a control channel. In the client receiving the data steam, any one of three situations might occur, two of which affect the remote cache server 124: (1) the client may recognize from the URL fingerprint and the content fingerprint that it already has the document; or (2) the application 104 requests the document being sent. The third situation is the absence of the other two (i.e., the document was not present and is not requested during its loading). If the client recognizes that it already has the document, it will communicate an abort to the remote cache server 124 (1710), which will then terminate the document transmission (1714). On the other hand, if the client cache assistant 106 indicates that the document being sent is now being requested (1716), then the remote cache server 124 will increase the priority of the transmission of the document (1720). The recognition of the requested document as being the same as the document being downloaded to the client could be done by either the client cache assistant 106 or the remote cache server 124.
The communication channels between the client cache assistant 106 and the remote cache server 124 allow for multiple data streams of varying priorities. One or more control channels may be used to allow the client cache assistant 106 and the remote cache server 124 to exchange control information such as abort or priority information. Different resources being downloaded to the client could be assigned different priorities based on any number of factors. Those of ordinary skill in the art would recognize various way to use prioritization.
Although some of various drawings illustrate a number of logical stages in a particular order, stages which are not order dependent may be reordered and other stages may be combined or broken out. While some reordering or other groupings are specifically mentioned, others will be obvious to those of ordinary skill in the art and so do not present an exhaustive list of alternatives. Moreover, it should be recognized that the stages could be implemented in hardware, firmware, software or any combination thereof.
The foregoing description, for purpose of explanation, has been described with reference to specific embodiments. However, the illustrative discussions above are not intended to be exhaustive or to limit the invention to the precise forms disclosed. Many modifications and variations are possible in view of the above teachings.
This application is a continuation of U.S. patent application Ser. No. 10/882,796, filed Jun. 30, 2004, entitled “System and Method of Accessing a Document Efficiently Through Multi-Tier Web Caching,” which is incorporated by reference herein in its entirety. This application is also related to U.S. patent application Ser. No. 10/882,794, filed on Jun. 30, 2004, entitled “System and Method of Accessing a Document Efficiently Through Multi-Tier Web Caching,” which is hereby incorporated by reference in its entirety.
Number | Date | Country | |
---|---|---|---|
Parent | 10882796 | Jun 2004 | US |
Child | 13550534 | US |