DETAILED DESCRIPTION OF THE INVENTION

[0029] Referring to FIG. 1, a wide area network 100 is shown. In the network 100, a client computer 101 can communicate with a server computer 102 by sending data over links 103 and 104 to a data network 130. The data network 130 may include multiple nodes 131-134 that can route data between the client 101 and the server 102. The client computer 1 01 may transmit and receive data using the TCP/IP, HTTP, and other protocols. For example, the client 101 may use the HTTP protocol to request web pages from the server 102.

[0030] Web pages and multimedia data sent from the server 102 to the client 101 may have a natural linear sequence associated with them. The natural sequence of video data may be the linear order of video frames while the natural sequence of text may be the order in which pages of text are arranged in a document. Data having a natural linear sequence can be streamed from a server to a client to minimize download delays. In a streaming system, while earlier items in a liner sequence are being processed and/or displayed, subsequent items may be downloaded to the client computer. When processing and/or display of an item is complete, processing or display or a fully received “streamed” item may quickly begin. Since receipt of a streamed item is fully or partially complete when the item is requested, a user or client application requesting the streamed item will perceive a reduced downloading delay. For example, if the first page of a document is retrieved by a user, the second page can be downloaded while the first page is being read. If the user continues reading at the second page of the document, that page will then be available at the client, such as in a cache area on a hard disk drive, and can be read without additional downloading delay.

[0031] Software execution may not follow a predictable natural linear order. Software may include jump statements, break statements, procedure calls, and other programming constructs that cause abrupt transfers of execution among sections of executing code. The execution path that is traversed during the processing of interrelated code modules (such as code segments, code classes, applets, procedures, and code libraries), will often be non- linear, user dependent, may change with each execution of the application program, and may change depending on the state of various data items. Although a natural order may be lacking, an advantageous order may be determined in which to stream modules. The order may be determined using criteria that is independent of the computer's internal architecture or internal operating system (execution environment) considerations.

[0032] Referring to FIG. 2, a software application 200 may include multiple modules “A” through “H.” Modules “A” through “H” may be Java Classes, C++ procedure libraries, or other code modules that can be stored at a server. Some of the modules “A” through “H” may also be stored at the client computer, such as in a hard disk drive cache or as part of a software library stored at the client computer. When a client computer begins execution of the application 200, a first module, such as module “A,” may be downloaded from the server and its execution at the client 410 may begin. As module “A” is being processed, the programming statements contained therein may branch to, for example, module “E.” If Module “E” is not already resident at the client, the execution of module “A” can be suspended, module “E” can be retrieved from the server, and then the execution of module “E” code may begin. In such a scenario, a user will experience a module download delay associated with retrieving module “E” from the server.

[0033] To minimize module download delays experienced by a user, module “E” may be transparently streamed from a server to the client computer. Transparent streaming allows future module use to be predicted and modules to be downloaded while other interrelated modules “A” are executing. Referring to FIG. 4, an exemplary software architecture 400 providing transparent streaming is shown. The software architecture 400 includes a server 401 having a database 403 of stored software modules. The server 401 can transparently transmit a stream of software modules 405 over a communications link to a client computer 410. The communication link may be an analog modem connection, a digital subscriber line connection, a local area network connection, or any other type of data connection between the server 401 and client 410. As particular software modules are being executed at the client 410, additional modules are sent from the server 401 to the client 410. In a dynamic streaming implementation, the order in which modules are streamed between the server and client may be altered based on the particular client computer 410 being served, based on the user of the client computer, and based on other dynamically determined factors.

[0034] Referring to FIG. 3, the execution order of application modules “A” through “H” may resemble a directed graph 300 rather than a linear sequence of modules. For example, as illustrated by the graph 300, after module “A” is executed, execution can continue at module “B,” “D,” or “E.” After module “B” is executed, execution can continue at module “C” or “G.” The execution path may subsequently flow to additional modules and may return to earlier executed modules.

[0035] The server 401 can use streaming control information 402 to determine the order in which to stream modules from the server 401 to the client 410. The streaming control information 402 can include, for example, a predicted execution flow between software modules such as that represented by the directed graph 300. As downloaded modules are executed by the client 410, the client may send control data 415 to the server 401 to dynamically update and alter the order in which modules are streamed from the server 401 to the client 410. Control data 415 may be used to request particular modules from the server 401, to send data regarding the current execution state of the application program, to detail the current inventory of modules residing in the client's local storage 411, and to report user input selections, program execution statistics, and other data derived regarding the client computer 410 and its executing software.

[0036] The sequence of modules sent in the stream 405 from the server 401 to the client 410 can be determined using a streaming control file 402. The streaming control file 402 includes data used by the server to predict modules that will be needed at the client 410. In a graph-based implementation, the control file 402 may represent modules as nodes of a directed graph. The control file 402 may also represent possible execution transitions between the modules as vertices (“edges”) interconnecting the nodes. Referring to Table 1, in a weighted graph implementation, the streaming control file 402 may include a list of vertices represent possible transitions between modules. For example, Table 1 list vertices representing all possible transitions between the modules “A” through “H” of graph 300 (FIG. 3). Each vertex in Table 1 includes a weight value indicating the relative likelihood that the particular transitions between modules will occur. In the example of Table 1, higher weight values indicate less likely transitions. The server 401 may apply a shortest-path graph traversal algorithm (also known as a “least cost” algorithm) to determine a desirable module streaming sequence based on the currently executing module. Example shortest-path algorithms may be found in Telecommunications Networks: Protocols, Modeling and Analysis, Mischa Schwartz, Addison Wesley, 1987, §6. 1

[0037] For example, Table 2 represents the minimum path weight between module “A” and the remaining modules of Table 1. 2

[0038] Based on the weight values shown in Table 2, the server 401 may determine that, during the execution of module “A”, the module streaming sequence “B,” “C,” “E,” “G,” “H,” “D,” “F” is advantageous. If a particular module in a determined sequence is already present at the client 402, as may have been reported by control data 415, the server 401 may eliminate that module from the stream of modules 405. If, during the transmission of the sequence “B,” “C,” “E,” “G” “H,” “D,” “F,” execution of module “A” completes and execution of another module begins, the server may interrupt the delivery of the sequence “B,” “C,” “E,” “G,” “H,” “D,” “F,” calculate a new sequence based on the now executing module, and resume streaming based on the newly calculated streaming sequence. For example, if execution transitions to module “B” from module “A,” control data 415 may be sent from the client 410 to the server 401 indicating that module “B” is the currently executing module. If module “B” is not already available at the client 410, the server 401 will complete delivery of module “B” to the client and determine a new module streaming sequence. By applying a shortest-path routing algorithm to the edges of Table 1 based on module “B” as the starting point, the minimum path weights between module “B” and other modules of the graph 300 (FIG. 3) can be determined, as shown in Table 3. 3

[0039] Based on the shortest path weights shown in Table 3, the server 401 may determine that module streaming sequence “C,” “G,” “E,” and “H” is advantageous.

[0040] Other algorithms may also be used to determine a module streaming sequence. For example, a weighted graph 300 may be used wherein heavier weighted edges indicate a preferred path among modules represented in the graph. In Table 4, higher assigned weight values indicate preferred transitions between modules. For example, edges (A,B), (A,D), and (A,E) are three possible transitions from module A. Since edge (A,B) has a higher weight value then edges (A,D) and (A,E) it is favored and therefore, given module “A” as a starting point, streaming of module “B” before modules “D” or “E” may be preferred. Edge weight values can be, for example, a historical count of the number of times that a particular module 5 was requested by a client, the relative transmission time of the code module, or a value empirically determined by a system administrator and stored in a table 402 at the server 401. Other edge weight calculation methods may also be used. 4

[0041] In an preferred-path (heavy weighted edge first) implementation, edges in the graph 300 having higher weight values are favored. The following exemplary algorithm may be used to determine a module streaming sequence in a preferred-path implementation:

[0042] 1: Create two empty ordered sets:

[0043] i) A candidate set storing pairs (S,W) wherein “S” is a node identifier and “W” is a weight of an edge that may be traversed to reach node “S.”

[0044] ii) A stream set to store a determined stream of code modules.

[0045] 2: Let Si be the starting node.

[0046] 3: Append the node S₁ to the Stream Set and remove any pair (S₁, W) from the candidate set.

[0047] 4: For each node S_j that may be reached from node S_i by an edge (S_i,S_j) having weight W_j:

[0048] {

[0049] If S_j is not a member of the stream set then add the pair (S_j, W_j) to the candidate set.

[0050] If S_j appears in more than one pair in the candidate set, remove all but the greatest-weight (S_j, W) pair from the candidate set.

[0051] }

[0052] 5: If the Candidate set is not empty Select the greatest weight pair (S_k, W_k) from the candidate set.

[0053] Let S_i=S_k

[0054] Repeat at step 3

[0055] For example, as shown in Table 5, starting at node “A” and applying the foregoing algorithm to the edges of Table 4 produces the stream set {A, B, C, E, H, G, D, F}: 5

[0056] Implementations may select alternative algorithms to calculate stream sets.

[0057] Application streaming may also be used to stream subsections of an application or module. For example, subsections of compiled applications, such as applications written in C, C++, Fortran, Pascal, or Assembly language may be streamed from a server 401 to a client 410. Referring to FIG. 5A, an application 500 may include multiple code modules such as a main code module 501 and code libraries 510 and 515. The main module 501 contains program code that is executed when the application is started. The code libraries 510 and 515 may contain header data 511 and 516 as well as executable procedures 512-514 and 517-519 that are directly or indirectly called from the main module 501 and other library procedures.

[0058] In a Microsoft Windows 95 / Microsoft Visual C++implementation, the main code module 501 may contain a compiled C++ “main” procedure and the library modules 510 and 515 may be dynamic link libraries having compiled C++ object code procedures. Header data 511 and 516 may include symbolic names used by operating system link procedures to dynamically link libraries 510 and 515 with the main module 501. Header data may also indicate the location of each procedure within the library. In a jump table implementation, a calling procedure may access library procedures 512-514, 517-519 by jumping to a predetermined location in the header 511 or 516 and from there, accessing additional code and/or data resulting in a subsequent jump to the start of the procedure.

[0059] Data and procedures within an application's code modules and libraries may be many hundreds or thousands of bytes long. Prior to executing an application, a client may need to retrieve a lengthy set of modules and libraries. By reducing the size of the module and library set, the initial delay experienced prior to application execution can be reduced. In a streaming implementation of application 500, code within subsections of the application's code modules can be removed and replaced by shortened streaming “stub” procedures. The replacement of application code with streaming stub procedures may reduce module size and associated transmission delay. For example, referring to FIGS. 5A and 5B, the code library 510 may include a header 511 that is 4 kilobytes (Kbytes) in length and procedures 512-514 that are, respectively, 32 Kbytes, 16 Kbytes, and 8 Kbytes. Referring to FIGS. 5B and 5C, to reduce the size of the library 510, procedures code 512-514 may be removed from the library 510 and stored in a streaming code module database 403 at the server 401 (FIG. 4). The removed procedure code 512-514 may be replaced by “stub” procedures 522-524 resulting in reduced-size code library 530 that can be linked with application modules 501 and 515 in place of library 510. Header data 511 of library 530 can include updated jump or link information allowing stub procedures 522-524 to act as link-time substitutes for procedures 512-514.

[0060] A server 401 may provide a streaming-enabled version of application 500 to a client 410 by sending main module 501, library module 515, “streamed” library 530, and, in some implementations, a streaming support file 535 to the client 410 in response to a request for the application 500. The streaming support file 535 may include procedures accessed by the stubs 522-524 to facilitate code streaming between the server 401 and client 410. At the client 410, modules 501, 515, 530 and 535 can be linked and execution of the resulting application can begin. As the main module 501 and various called procedures are executed at the client 410, code modules stored in the database 403 can be streamed from the server 401 to the client 410. Data may be included in the stream 403 to identify stub procedures 522-524 associated with the streamed code modules. As the streamed modules are received at the client, they are integrated with the executing application.

[0061] In an appended module implementation, streamed code modules are integrated with the executing application by appending received modules to their corresponding library or code file. For example, referring to FIGS. 5C and 5D, as modules 512-514 are streamed from the server to the client, they are appended to the library file 530 thereby forming an augmented library file 540. As the modules 512-514 are streamed from the server 401 and appended to the file 530, header data 511 or stub data 522-524 is updated so that the now-appended modules are accessible from a calling procedure. For example, referring to FIG. 5D, an additional “jump” may be added between each stub procedure 522-524 and its associated appended module 512-514. Alternatively, header data 511 may be updated so that procedures 512-514 are accessible in place of stubs 522-524. In a stub-replacement implementation, stubs 522-524 are replaced by procedure modules 512-514 as the modules are received from the server 401. Stub replacement may require altering or rearranging the location of the remaining stubs or procedures within a code module or library as replacement code is received. Implementations may employ still other methods of integrating streamed code with executing applications and modules.

[0062] In some scenarios, removed code, such as procedure code 512-514 which, in the example given, was replaced by stubs 522-524, may be required (called by another procedure) before it is streamed from the server 401 and integrated with the module 530. In such a case, stub code 522-524 may access streaming functions in the streaming support library 535 to obtain the required procedure. To do so, the streaming support library 535 may send control data 415 to the server 401 to request the needed procedure. In response, the server 401 can halt the current module stream 405 and send the requested module. Upon receipt of the requested module, procedures in the streaming support library 535 may be used to integrate the received module with the application and to continue with the execution of the requested module. The server may thereafter determine a new module stream based on the requested module or other control data 415 that was received from the client.

[0063] Code modules may be reduced in size without the use of stub procedures. For example, referring again to FIGS. 4, 5A, 5B, and 5E, in a interrupt driven implementation, procedure code 512-514 may be removed from a code library 510 and stored in a database 403. Header information 511 as well as data indicating the size and location of removed procedure code 512-514 may then be transmitted to a client 410. The client 410 may construct a new library 550 by appending a series of interrupt statements in place of the removed procedure code 512-514. When the application 500 is executed, the code library 550 is substituted for the library 510 and execution of the program 500 may begin. As the program 500 executes, the removed procedure code 512-514 can be streamed to the client 410 and stored in a local database 411. If the application 500 attempts to execute procedure code 512-514 it may instead execute one of the interrupt statement that have replaced procedure code 512-514. The execution of the interrupt statement halts the execution of the program 500 and transfers control to a streaming executor program 416.

[0064] Executor 416 implements interface technology similar to that of a conventional run-time object code debugger thereby allowing the executor 416 to intercept and process the interrupt generated by the application 500. When the interrupt is intercepted by the executor 416, data provided to the executor 416 as part of the client execution platform (operating system) interrupt handling functionality can be used to identify the module 550 in which the interrupt was executed and the address of the interrupt code within the module. The executor 416 then determines whether procedure code 512-514 associated with the interrupt location has been received as part of the module stream 405 sent to the client. If the appropriate procedure code has been received, the executor 515 replaces the identified interrupt with the its corresponding code. For example, procedures 512-514 may be segmented into 4 Kilobyte code modules that are streamed to the client 410. When an interrupt statement is executed by the application 500, the executor 416 intercepts the interrupt, determines an appropriate 4 Kilobyte code block that includes the interrupt statement, and replaces the determined code block with a received code module. If the appropriate code module has not yet been received, an explicit request may be sent from the client 410 to the server 401 to retrieve the code module prior to its insertion in the library 550. The executor 416 may thereafter cause the application 500 to resume at the address of the encountered interrupt.

[0065] Implementations may also stream entire modules or libraries. For example, main code module 501 may be received from the server 401 and begin execution at the client 410 while code libraries 510 and 515 are streamed from the server 401 to the client 410. Integration of streamed modules with executing modules may be provided by client 410 dynamic module linking facilities. For example, delay import loading provided by Microsoft Visual C++ 6.0 may be used to integrate streamed modules 510 and 515 with executing modules 501. Dynamic linking of streamed modules may be facilitated by storing the streamed modules on a local hard disk drive or other storage location accessible by client 410 link loading facilities. In an exemplary implementation, streaming is facilitated by altering client 410 operating system link facilities such that the link facility can send control data 415 to the server 401 to request a particular module if the module is has not already been streamed to the client 401.

[0066] In a protected-memory computer system, direct manipulation of executing application code and data may be restricted. In such systems, a “kernel” level processes or procedure may be required to support integration of streamed modules with executing application. In such a case, streaming support 535 may be pre-provisioned by installing support procedures at the client 410 prior to the client's request for the application 500.

[0067] Archive file such as CAB, tar, BINHEX, and ZIP files, including ZIP formatted Java Archive (JAR) files, can be streamed by extracting modules of data from the archive files and streaming those modules to a client terminal. In the following text, streaming of a Java Archive is detailed. The disclosed techniques may be applied to the streaming of other archive formats as well.

[0068] Java classes, as well as images, sounds, and other data files, can be aggregated together in a Java Archive files (a “JAR file”). A JAR file is a data archive using the ZIP archive format to aggregate a collection of logically separate data files.. The files in a JAR file can include Java class files, video, sound, and other data files, as well as meta-information files (the contents of which are define by Sun Microsystems Inc.'s JAR standard).

[0069] FIG. 6 shows the format of an example JAR file 600. The content of elements 611-616, 621-626, 641-647 of the file 600 is are fields defined by the ZIP specification (this specification is available from PKWARE, Inc.) These elements 611-616, 621-626, 641-647 organize and support the archiving and compression of files 601-606 stored in the archive 600. Elements 611-616 of the archive 600 are known as local file headers. Each local file header precedes a file stored in the archive 600. Local file headers specify, among other things, the name and relative path of the associated file, the last time/date the file was modified, and the compressed and uncompressed size of the file. The contents of each local file header is repeated (together with additional information) in a central directory 640 located at the end of the ZIP archive. For each local file header 611-616 in the archive, the central directory 640 includes a corresponding central directory headers 641-646. According to the ZIP format specification, each file 601-606 in the archive may also include an optional data descriptor 621-626; though these data descriptors are typically not present in JAR files.

[0070] A ZIP file's central directory 640 includes an ‘end of central directory’ record 647. The record 647 includes, among other things, the total number of entries in the central directory (i.e., the total number of files in the archive), the size of the central directory, the number of the disk containing the central directory, and the offset of the start of the central directory with respect to the start of an archive segment containing the central directory 640 (a ZIP archive may be split among multiple segments, e.g., multiple floppy disk of files).

[0071] The size of a ZIP archive 600 can be obtained in a number of different ways. One method is to add the compressed sizes of the files 601-606 (obtained from the compressed file size information in either the local file header or central directory header), together with the sizes of the headers 611-616, 621-626, 641-647. A second method (usable if the ZIP archive is contained within one file or disk, as is the case with JAR files), determines the size of the ZIP archive from information in the end of central directory record 647. In particular, the value in a ‘size of the central directory’ field of the record 647 is added to the value of the ‘offset of start of central directory with respect to the starting disk number’ field in the record 647.

[0072] Files 601-606 are the files being archived using the ZIP compression technique and format. Files 601-606 include files used by an executing application 603-606 as well as meta-information files 601-602 containing information about one or more of the other files 603-606. The meta-information includes a “manifest” file 601. Conventionally, this manifest file is named “MANIFEST.MF” and is logically associated with the storage path “META-INF” (i.e., the file has a relative path/name “META-INF/MANIFEST.MF”). The manifest file 601 includes, among other things, a JAR format version number and the names of the files 603-606 that are in the JAR 600. A manifest file also may includes optional “digest” data for one or more of the files 603-606. Digest data is generated by processing a file (e.g., file 603) with a cryptography algorithm (typically, the MD5 or SHA algorithm). The digest data is stored in the manifest file and serves as a digital signature that can be used to validate the file 603 (i.e., to ensure that the file has not been modified or corrupted). In addition to file validation, the digest information, along with separately stored signature information 602, can be used to identify a file's signer. Signature files are another form of meta-information that can be stored in a JAR. Java Archives are more fully described in, e.g., the text Java in a Nutshell, by David Flanagan, O'Reilly, 1997. In resource limited devices, such as J2ME/CLDC/MIDP devices, simplified meta-information, excluding signature files 602, is used.

[0073] Conventionally, a JAR file 600 is transmitted as a single file from a server 401 to a client device 410. However, the streaming server 401 may stream the file 600 as a series of separate modules by extracting the ZIP central directory 640, as well as the individual files and their associated headers 631-636, and streaming each of these elements 631-636, 640 separately. Typically, the central directory information 640 will be streamed first, followed by the meta-information 631-632 (if present), and then the remaining modules 633-636. In some implementations, the modules 631-636 or a subset, e.g., modules 633-636, will be ordered by a predictive algorithm, however such ordering is not required in all implementations. Predictive ordering may be applied to a subset 633-636 of the modules and a fixed ordering to other modules 631-632, 640.

[0074] Referring back to FIG. 4, to stream a JAR file 600, the server 401 extracts the contents of the JAR file as a set of streamable modules 631-636, 640 (other divisions and combinations of the file 600 can be used; for example, each element 601-606, 611-616, 621-626, 641-647 may be a separate module). To extract the modules 631-636, 640, their size and location is determined based on the offset of the module within the file 600, the compressed size of the files 601-606, the size each local header 611-616 and the size of each data descriptor 621-626 (if present). This size and offset information can be determined from the central directory file headers 641-647. The extracted modules 631-636, 640 are then ordered for streaming to the client 410.

[0075] The receiving device 410 includes a streaming executor 416 that controls the receipt of the streamed modules 631-636, 640 and the storage of those modules in storage 411 (which may be, e.g., solid state memory such as RAM or EEPROM, or magnetic memory such as a hard disk). The stored modules may then be accessed by a Java application or applet. The component 416 can incrementally rebuild a JAR file at the receiving device 410 to store the received modules. For example, a cell phone, PDA, or other device supporting the Java Connected Limited Device Configuration (CLDC) on the Java 2 Micro Edition (J2ME) architecture, may store the received modules in an incrementally rebuilt JAR file in memory 411. Storing the received modules in a JAR file can provide for more efficient use of limited memory resources.

[0076] To reconstruct the JAR file 600, the receiving device 410 may pre-allocate space for the file 600. The size of the archive can be determined from the end of central directory record 647. To enable space allocation for the file 600, streaming of the modules 631-636, 640 may begin with streaming of the central directory module 640 (or, at least, of the end of central directory record 647). Upon receipt of the record 647, the client device can process the size and offset information in the record 647 to determine the total size of the JAR archive 600. The client may then allocate space for the JAR archive 600. Alternatively, control data 415 can be sent to the client 410 identifying the JAR archive to be streamed and the size of the archive.

[0077] After allocating space for the JAR archive, the client places the module 640 in the allocated space at the proper offset within that space (as indicated by the offset information in the record 647). The remaining modules 631-636 may then be streamed from the server 401 to the client 410. In some implementations, streaming of the modules 631-636 may be in accordance with a predictive algorithm. As the modules 631-636 are received at the client 410 from the server 401, they can be consecutively placed in the space allocated for the JAR file. In some cases, this may result in a different ordering of the modules 631-636 in the reconstructed JAR file than the ordering of the modules in the original JAR file. For example, if module 636 is streamed before modules 631-635, then module 636 will be located at the beginning of the reconstructed JAR archive and the remaining modules 631-635 thereafter. If such a consecutive placement of received modules in the reconstructed JAR file is used, resulting in a different order of modules 631-636 in the reconstructed file than in the original file, then the ‘relative offset of local header data’ contained in the module's corresponding central directory header 641-646 will need to be adjusted to reflect the new placement of each module. Alternatively, the ‘relative offset of local header data’ can be used to position modules 631-636 in their original position regardless of the streaming order of the modules 631-636.

[0078] Java programs executing at the client may begin using a module 633-636 following the receipt of that module; that is, the client does not need to receive all of the modules 631-636, 640 before received modules can be used. Instead, data in the fully or partially reconstructed JAR file may be extracted and de-compressed as needed using decompression and file management functionality present at the receiving device.

[0079] The client device's file extraction code may be modified to generate an exception if a module requested by an executing application is missing from the archive. Thus, if the Java application request file 616 from the archive 600, and that file has not yet been received and placed in the client's reconstructed archive, the extraction software can generate an exception. In response to this exception (which may be implemented as a procedure call to executor 416 routines), the executor 416 sends control data 415 to the server 401 identifying the requested module or file (i.e., module 636 or file 606). The server 401 may then change the streaming order of modules so that the requested module is streamed next. In some cases, a currently streaming module may be interrupted to allow immediate transmission of the requested module. This may result in discarding of the partially received module at the client. Alternatively, the requested module may be sent after any partially streamed modules. Interruption of a partially streamed module may be determined based on a weighing algorithm (e.g., finish if greater than 50% already transmitted, discard otherwise).

[0080] Java implementations may require the presence of the manifest file 601 at the client before any of the files 603-606 can be used. In such implementations, it is desirable to stream the manifest file 601 before the files 603-606. Similarly, signature files 602, if used, should be streamed prior to, or immediately following, the respective signed file 603. Streaming of the manifest file 601 and other meta information 602 in this manner permits checking and validation of the manifest information as well as the processing of digital signature information as the files 603-606 are received.

[0081] In some implementations, redundant information in an archive file may be omitted prior to streaming. For example, the local file header data 611-616 is repeated in the central directory headers 641-646; consequently, the local file header data 611-616 may be discarded from the streamed modules (resulting in a smaller module), and can be reconstructed at the client using the data in the central directory headers 641-646.

[0082] For some data types, improved compression may be obtained using type-specific compression algorithms (rather than standard ZIP compression algorithms). Prior to streaming a module, the module may be decompressed and re-compressed into a new compression format. In such cases, additional software functionality (i.e., additional decompression algorithms and ZIP compression) may be implemented at the receiving device for the reconstruction of the JAR file.

[0083] If the receiving device 410 is not resource constrained, then the streamed files also may be stored at the receiving device in a de-archived format. For example, each received file may be stored as a separate files on a hard disk 411 (or other type of memory 411). Path information in the local or central directory file headers 611-616, 641-646, or sent in control data 415, can be used to define a hierarchical storage of the streamed files at the receiving device.

[0084] In some cases, an extracted file may be further subdivided before streaming. For example, file 601 may be a class file containing several Java classes. Individual classes may be extracted from the file 601 and streamed as separate modules. This extraction of classes from within a file 601 is an implementation of the technique discussed with respect to FIGS. 5A-5E.

[0085] Other methods of determining stream sets may be used. In a list-based implementation, the streaming control file may include predetermined list of module streaming sequences. For example, the streaming control file 402 may include a module streaming sequence list associated with a first user and a second module streaming sequence list associated with a second user. Control data 415 sent from the client 410 to the server 401 may identify the current user at the client 410. Once the user has been identified to the server, the server may stream software modules in accordance with the user's associated streaming sequence list. User-based streaming data may be advantageous where a user's past behavior can be used to anticipate the order of modules to be accessed by that user.

[0086] In graph-based streaming control file implementations, the weight of edges connecting nodes may be determined statically or dynamically and may be determined based on a collection of historical usage data. For example, in a programmer-controlled implementation, a software programmer estimate the likelihood that particular transitions between nodes will occur based on the programmer's knowledge of the software code and the expected application usage patterns. Alternatively, application profiling programs may be used to gather run-time execution data recording transitions between various applets, Classes or code modules and thereby determine the likelihood that particular transitions will occur. In a client-feedback implementation, control data 415 sent from the client 410 to the server 401 during module execution is used to build a statistical database of module usage and, based on that database, determine the module streaming order.

[0087] In a client-controlled streaming implementation, streaming control data 402 may be located at the client 410 and control data 415 sent from the client 410 to the server 401 may be used to sequentially request a stream of modules from the server. For example, while the client computer 410 is executing a first module, a background process may send control data 415 to a server to request additional modules that can be buffered on a hard disk 411 at the client computer 410. A client-controlled streaming implementation may used existing HTTP servers and HTTP protocols to send request from the client 410 to the server 401 and send software modules from the server 401 to the client 410. Furthermore, although streaming of software modules has been emphasized in the foregoing description, non-executable data, such as hypertext markup language, binary graphic files, and text, may be streamed as a collection of modules.

[0088] Implementations may include a “handshaking” procedure whereby, at the start of application execution, control data 415 is sent between the server 401 and the client 410. The handshaking data may include an inventory of application modules residing at the client and at the server. Such handshaking data allows both the client 410 and server 401 to determine their respective software module inventory and to optimize the stream of software modules based on that inventory information.

[0089] In a history-dependent implementation, a server or client can store data about a series of transitions between modules and calculate a new module stream based on a history of transitions. For example, referring to FIG. 3, if the module “G” was reached by the path A-B-G, then a server or client may determine that module “E” followed by “H” is to be streamed. On the other hand, if the module “G” was reached by the path A-B-C-G then the streaming sequence may include only the module “H.”

[0090] The invention may be implemented in computer hardware, firmware, software, digital electronic circuitry or in combinations of them. Apparatus of the invention may 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 may be performed by a programmable processor executing a program of instructions to perform functions of the invention by operating on input data and generating output.

[0091] The invention may advantageously be implemented 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 may 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 may 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.

[0092] 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 may be supplemented by, or incorporated in, specially-designed ASICs (application-specific integrated circuits).