System and method for dynamically and persistently tracking incremental profiling data in a process cloning application environment

<u>Field of the Invention</u>

0001The invention relates in general to tracking profiling data and, in particular, to a system and method for dynamically and persistently tracking incremental profiling data in a process cloning application environment.

<u>Background of the Invention</u>

0002Recent advances in microprocessor design and component integration have enabled a wide range of devices to offer increasingly complex functionality and "soft" features. Soft features include software applications that enhance and customize the operation of a device. These devices include standard computing devices, such as desktop and laptop computers, portable computing devices, such as personal data assistants, and consumer devices, such as cellular telephones, messaging pagers, gaming consoles, and set top boxes. Most devices now include an operating system to support soft features and other extensions.

0003The increased capabilities offered by these software-upgradeable devices have also created certain user expectations. Often, users are not technically savvy and are intolerant of performance compromises occasioned by architectural challenges, such as slow or inconsistent application performance. Similarly, users generally expect to be able to access a host of separate applications, which are implemented at the system level through multitasking. For users, widely available software applications ensure a positive experience through consistency and increased exposure across multiple platforms. However, for software developers, engineering software applications for disparate computing platforms entails increased development costs and on-going support and upgrade commitments for each supported architecture.

0004Managed code platforms provide one solution to software developers seeking to support multiple platforms by presenting a machine-independent and architecture-neutral operating environment. Managed code platforms include programming language compilers and interpreters executed by an operating system as user applications, but which provide virtual runtime environments within which compatible applications can operate. For instance, applications written in the Java programming language, when combined with a Java virtual machine (JVM) runtime environment, can operate on heterogeneous computer systems independent of machine-specific environment and configuration settings. An overview of the Java programming language is described in P. van der Linden, "Just Java," Ch. 1, Sun Microsystems, Inc. (2d ed. 1997), the disclosure of which is incorporated by reference. JVMs are a critical component to the overall Java operating environment, which can be ported to the full range of computational devices, including memory-constrained consumer devices.

0005Managed code platforms are generally designed for the monotonic execution of a single application instance. Multiple instances of a managed code platform are executed to simulate multitasking behavior. Such forced concurrency, however, creates several performance problems. First, each instance incurs a startup transient. Executable and startup data must be read from slow persistent storage, which results in slow initial application performance. Similarly, memory is not shared between instances and each additional instance increases the overall memory footprint of the platform by separately loading and instantiating classes, generally problematic in memory-constrained systems. Moreover, data dependencies and deferred initialization of system state can result in non-deterministic execution patterns. Finally, each instance independently determines the relative importance of executing methods and compiles machine code on an <i>ad hoc</i> basis, often causing inconsistent application performance.

0006One solution to providing concurrently executable managed code platforms involves designating a managed code platform that executes as a master process. Individual child managed code platform processes are spawned from the master process using process cloning and inherit runtime state from the master managed code platform. As a result, each child process benefits from the initialization actions completed by the master process.

0007Accurately selecting the initialization actions is important since the inherited runtime state directly affects the quality of memory sharing and savings, startup and warmup times saved by each child process, and whether execution determinism is improved. In one approach, initialization actions are chosen to improve child process performance by initializing the runtime state based on <i>a priori</i> knowledge of likely execution profile. For instance, commonly used methods can be precompiled or class loaders preemptively invoked to build inheritable, warmup state into the master process context. However, child process performance could be adversely affected if the underlying execution profile is poorly matched to real runtime conditions. For example, unused methods could be precompiled or incorrect classes could be loaded, thereby consuming processing and memory resources in the child process, which must also compile the methods and load the classes actually used. Similarly, the device usage patterns could change. As a result, a single set of initialization actions could inaccurately reflect the preferable set of actions best suited for different usage and execution scenarios.

<u>Summary of the Invention</u>

0008In accordance with one embodiment of the invention, a managed code platform is executed in an application framework that supports the spawning of multiple and independent isolated user applications. Preferably, the application framework supports the cloning of the memory space of each user application using copy-on-write semantics. The managed code platform includes a master runtime system process, such as a virtual machine, to interpret machine-portable code defining compatible applications. An application manager also executes within the application framework and is communicatively interfaced to the master runtime system process through an inter-process communication mechanism. The application framework logically copies the master runtime system process context upon request by the application framework to create a child runtime system process through process cloning. The context of the master runtime system process stored in memory is inherited by the child runtime system process as a prewarmed state and cached code. When implemented with copy-on-write semantics, the process cloning creates a logical copy of references to the master runtime system process context. Segments of the referenced master runtime system process context are lazily copied only upon an attempt by the child runtime system process to modify the referenced context. At runtime, each child runtime system process profiles an application to create incremental profile data, which can be transiently staged in a memory buffer or persistently stored in a file. The application manager retrieves and analyzes the incremental profile data and updates a master warmup program. The master runtime system process executes the updated master warmup program either upon initialization or in response to a command from the application manager to dynamically reinitialize the context. An example of a suitable managed code platform and runtime system process are the Java operating environment and Java virtual machine (JVM) architecture, as licensed by Sun Microsystems, Inc., Palo Alto, CA.

0009One embodiment of the invention provides a system and method for dynamically and persistently tracking incremental profiling data in a process cloning application environment. A master runtime system process is executed. A memory space of the master runtime system process is cloned as a child runtime system process responsive to a process request. The child runtime system process is executed. The execution of the child runtime system process is profiled by collecting profiling data incrementally. The child runtime system process profiles are fed back to the master runtime system process to benefit subsequent cloned child runtime system processes. In a further embodiment, the child runtime system process profiles are maintained in a persistent storage for use by the master runtime system process upon the next start up.

0010The use of the process cloning mechanism provided by the underlying application framework provides several benefits in addition to resolving the need for efficient concurrent application execution of machine portable code. The inheritance of a prewarmed state through the cloning of the master runtime process context provides inter-process sharing of pre-computed, pre-warmed state. Similarly, each child runtime system process executes in isolation from each other process, thereby providing strong resource control through the system level services of the application framework. Isolation, reliable process invocation and termination, and resource reclamation are available and cleanly provided at an operating system level. In addition, process cloning provides fast user application initialization and deterministic runtime behavior, particularly for environments providing process cloning with copy-on-write semantics. Finally, for non-shareable segments of the master runtime system process context, actual copying is deferred until required through copy-on-write semantics, which avoids impacting application performance until, and if, the segment is required.

0011The present approach therefore profiles the execution and performance of cloned runtime system processes to measure actual usage patterns at runtime, particularly relative to prewarmed state inherited from a master runtime system process. Preferably, such an approach provides dynamically applicable feedback or persistently stored profile data, which can be executed by a master runtime system process at startup or runtime, thereby providing fine tuning either at startup or dynamically at runtime based on observed profile data.

<u>Brief Description of the Drawings</u>

0012Various embodiments of the invention will now be described in detail by way of example only with reference to the following drawings: <ul id="ul0001" list-style="none" compact="compact"><li>FIGURE 1 is a functional block diagram showing, by way of example, runtime environments implemented on a plurality of heterogeneous devices.</li><li>FIGURE 2 is a block diagram showing a system for dynamically and persistently tracking incremental profiling data in a process cloning application environment, in accordance with one embodiment of the invention.</li><li>FIGURE 3 is a block diagram showing, by way of example, a master JVM process mapped into memory.</li><li>FIGURE 4 is a block diagram showing, by way of example, a master JVM process and a cloned JVM process mapped into memory through memory space cloning.</li><li>FIGURES 5A-B are block diagrams showing, by way of example, a master JVM process and a cloned JVM process mapped into memory through memory space cloning with copy-on-write semantics.</li><li>FIGURE 6 is a flow diagram showing a method for dynamically and persistently tracking incremental profiling data in a process cloning application environment, in accordance with one embodiment of the invention.</li><li>FIGURE 7 is a flow diagram showing the routine for loading a master JVM process for use in the method of FIGURE 6.</li><li>FIGURE 8 is a flow diagram showing the routine for cloning a process without copy-on-write semantics for use in the method of FIGURE 6.</li><li>FIGURE 9 is a flow diagram showing the routine for cloning a process with copy-on-write semantics for use in the method of FIGURE 6.</li></ul>

<u>Detailed Description</u>

System Overview

0013FIGURE 1 is a functional block diagram 10 showing, by way of example, runtime environments (RTEs) 14, 22, 24, 26 implemented on a plurality of heterogeneous devices 11. Each heterogeneous device 11 provides a managed code platform, such as the Java operating environment, executing in a runtime environment 14, 22, 24, 26, as further described below with reference to FIGURE 2. The heterogeneous devices 11 include, nonexclusively, a client computer system 13, such as a desktop or laptop computer system. Each client 13 is operatively coupled to a storage device 15 and maintains a set of classes 16 and class libraries 17, which respectively define code modules that specify data structures and sets of methods that operate on the data, and shareable collections of the modules. The heterogeneous devices 11 also include portable computing devices, including personal data assistants 21, and consumer devices, such as cellular telephones 23 and set top boxes (STB) 25. Finally, a server 18 is operatively coupled to a storage device 19 in which globally shareable class libraries 20 are maintained. Each of the heterogeneous devices 11 can interface via a network 12, which includes conventional hardwired and wireless network configurations. Other types of heterogeneous devices 11 and various network configurations, arrangements, and topologies are possible.

0014Each heterogeneous device 11 includes an operating system to manage resources, provide access to peripheral devices, allocate memory resources, and control program execution and termination. Each operating system supports a process cloning mechanism that spawns multiple and independent isolated user applications by cloning the memory space of specifiable processes. An example of a process cloning mechanism suitable for use herein is the <i>fork()</i> system call provided by the Unix or Linux operating systems, such as described in M.J. Bach, "The Design Of The Unix Operating System," Ch. 7, Bell Tele. Labs., Inc. (1986), the disclosure of which is incorporated by reference. The process invoking the <i>fork()</i> system call is known as the parent process and the newly created process is called the child process. The operating system assigns a separate process identifier to the child process, which executes as a separate process. The operating system also creates a logical copy of the context of the parent process by copying the memory space of the parent process into the memory space of the child process. In a copy-on-write variant of the <i>fork()</i> system call, the operating system only copies references to the memory space and defers actually copying individual memory space segments until, and if, the child process attempts to modify the referenced data of the parent process context. The copy-on-write <i>fork()</i> system call is faster than the non-copy-on-write <i>fork()</i> system call and implicitly shares any data not written into between the parent and child processes.

System for Dynamically Tracking Profiling Data Incrementally

0015FIGURE 2 is a block diagram 30 showing a system for dynamically and persistently tracking incremental profiling data in a process cloning application environment, in accordance with one embodiment of the invention. The system comprises a runtime environment 31 and individual classes 45 and class libraries 46 that form the overall core managed code platform. By way of example, the system is described with reference to the Java operating environment, although other forms of managed code platforms that execute applications preferably written in an object oriented programming language, such as the Java programming language, could also be used.

0016The illustrated runtime environment 31 includes an application manager 32, a master Java virtual machine (JVM) process 33 and zero or more cloned JVM processes 34. The master JVM process 33 and the cloned JVM processes 34 respectively correspond to a master runtime system process and child runtime system processes. The master runtime system process, preferably provided as a virtual machine, interprets machine-portable code defining compatible applications.

0017The runtime environment 31 executes an application framework that spawns multiple independent and isolated user application process instances by preferably cloning the memory space of a master runtime system process. An example of an application framework suitable for use herein is the Unix operating system, such as described generally in M.J. Bach, <i>supra</i> at Ch. 2.

0018The application manager 32 presents a user interface through which individual applications can be selected and executed. The application manager 32 and master JVM process 33 preferably communicate via an inter-process communication (IPC) mechanism, such as a pipe or a socket. The master JVM process 33 is started at device boot time.

0019Upon initialization, the master JVM process 33 reads an executable process image from the storage device 43 and performs bootstrapping operations specified in a master warmup program 44. The master warmup program 44 is an executable program performing a full range of operations and specifying warmup initialization actions, including preloading classes 45 and classes defined in the class libraries 46, precompiling methods, and executing static initialization methods, such as respectively described in commonly-assigned U.S. Patent application Serial No. 10/745,023, entitled "System And Method For Dynamic Preloading Of Classes Through Memory Space Cloning Of A Master Runtime System Process," filed 22 December 2003, pending; Serial No. 10/745,020, entitled "System And Method For Providing Precompiled Code Through Memory Space Cloning Of A Master Runtime System Process," filed 22 December 2003, pending; and Serial No. 10/745,021, entitled "System And Method For Eliminating Static Initialization Overhead By Memory Space Cloning Of A Master Runtime System Process," filed 22 December 2003, pending; the disclosures of which are incorporated by reference. Thus, upon completion of initialization, the memory image of the master JVM process 33 resembles that of an initialized, primed and warmed up JVM process with key classes, methods and initialized state stored in the master JVM process context.

0020Following the initialization, the master JVM process 33 idles, that is, "sleeps" in an inactive state, while awaiting further instructions from the application manager 32. The master JVM process 33 awakens in response to requests received from the application manager 32 to execute applications. The application manager 32 sends a request to the master JVM process 33, including standard command line parameters, such as application name, class path, and application arguments. The master JVM process 33 awakens and creates a cloned JVM process 34 as a new cloned process instance of the master JVM process 33 using the process cloning mechanism of the underlying operating system. The context of the master JVM process 33 stored in memory as prewarmed state 37 is inherited by the cloned JVM process 34 as inherited a prewarmed state 41, thereby saving initialization and runtime execution times and providing deterministic execution behavior. Following the "cloning" of the cloned JVM process 34, the master JVM process 33 records the launched application in an applications launched list 51 and returns to an inactive sleep state.

0021When implemented with copy-on-write semantics, the process cloning creates a logical copy of only the references to the master JVM process context. Segments of the referenced master JVM process context are lazily copied only upon an attempt by the cloned JVM process to modify the referenced context. Therefore, as long as the cloned JVM process does not write into a memory segment, the segment remains shared between parent and child processes.

0022The master JVM process 33 recognizes the following basic commands received from the application manager 32 through the IPC mechanism: <dl id="dl0001" compact="compact"><dt>(1) <i>list</i>:</dt><dd>Provides a list of applications launched in response to requests received from the application manager 32.</dd><dt>(2) <i>jexec</i>:</dt><dd>Invokes the process cloning mechanism, parses command line arguments and executes a new instance of the master JVM process 33 as the cloned JVM process 34. Preferably adopts a syntax compatible to standard JVM processes.</dd><dt>(3) <i>kill</i>:</dt><dd>Terminates the application identified by an application handle or process identifier.</dd><dt>(4) <i>source</i>:</dt><dd>Parses arguments and executes a program in the body of the master JVM process 33 without creating a cloned JVM process 34. Subsequently cloned JVM processes 34 inherit the refined memory state of the master JVM process 33 resulting from the execution of a <i>source</i> command on behalf of the master JVM process 33.</dd></dl>

Other commands are possible.

0023At device boot time, the master JVM process 33 executes the master warmup program 44 and the prewarmed state 37 and compiled method code in the code cache 38 are inherited by each cloned JVM process 34 as the inherited prewarmed state 41 and the inherited compiled method code in the inherited code cache 42. The master warmup program 44 is executed by the application manager 32 based on persistent profile data previously retrieved and stored from profile collectors 39 executing in one or more previously executed cloned JVM processes 34. Also, each profile collector 39 collects and stores incremental profile data generated by a profiler 40 executing in one or more cloned JVM processes 34. The dynamically profiled data is polled regularly by the application manager 32 to create a master dynamic profile 49, which is fed back into the master JVM process 33 using the <i>source</i> command to benefit subsequent cloned JVM processes 34. The master dynamic profile 49 also forms the basis of a persistent profile that can be used by the master JVM process 33 upon the next device boot.

0024Each profiler 40 performs accounting on applications being executed, as would be appreciated by one skilled in the art. The profiler 40 continually evaluates the application execution through, for example, an interpretive loop that attempts to identify important and performance sensitive code segments in the application. By way of example, the profiler 40 identifies method invocations and system library calls through the accounting, which includes counting loop iterations and method invocations and by sampling running methods. Other forms and approaches to performing accounting are possible. Preferably, the profiler 40 applies a threshold to the data collected through profiling and identifies those methods exceeding the threshold. The profile collector 39 evaluates the actions performed by the profiler 40 and stores those events affecting performance, such as class initializations and methods selected for dynamic compilation, as profile data. Other types and forms of profile data are possible.

0025In one embodiment, the profile collector 35 of each cloned JVM process 34 stores the collected profile data transiently in a memory buffer 52. Each cloned JVM process 34 maintains a profile request listener thread (not shown) executing concurrently to a main execution thread (also not shown). The application manager 32 accesses the profile request listener thread using a suitable IPC mechanism, such as a socket. At runtime, the cloned JVM process 34 listens to the profile thread through which the application manager 32 periodically sends profiling data queries, such polling requests. In response, the cloned JVM process 34 sends incremental profile data to the application manager 32 and resets the memory buffer 52. Only the profile data incrementally collected since the last profiling data query is staged in the memory buffer 52 and the application manager 32 stores assimilated sets of incremental profile data in the master dynamic profile 49.

0026In a further embodiment, the profile collector 35 of each cloned JVM process 34 stores the collected profile data persistently in a file as stored profiles 48. At runtime, the application manager 32 iterates over the incremental profile data in the stored profiles 48, preferably on a periodic basis, for events of interest and stores evaluated incremental profile data in the master dynamic profile 49. The incremental profile data can also be stored both as transient profile data in the memory buffer 52 and as persistent profile data in the stored profiles 48.

0027At runtime, the application manager 32 maintains the master dynamic profile 49, which assimilates the incremental profile data received from each cloned JVM process 32. The master dynamic profile 49 reflects dynamically observed events of interest, such as class initializations and methods selected for dynamic compilation. This data can be used to fine tune the warmup initialization actions executed by the master JVM process 33 upon initialization to reflect actual usage patterns at runtime. Thus, the application manager 32 analyzes the master dynamic profile 49 and updates the master warmup program 44. In one embodiment, the application manager 32 persistently stores the master warmup program 44 as a file in the storage device 43 at device shutdown or on a periodic basis, such as in response to an automatic save feature. Thereafter, at device boot time, the master JVM process 33 executes the saved master warmup program 44, which reflects the latest observed usage patterns. The prewarmed state 37 and compiled method code in the code cache 38 are then inherited by each cloned JVM process 34 as inherited prewarmed state 41 and inherited compiled method code in the inherited code cache 42.

0028In a further embodiment, the application manager 32 dynamically feeds the incremental profile data to the master JVM process 33. The master JVM process 33 recognizes an additional basic <i>source</i> command received from the application manager 32 through the IPC mechanism, such as described in commonly-assigned U.S. Patent application Serial No. 10/745,164, entitled "System And Method For Performing Incremental Initialization Of A Master Runtime System Process," filed 22 December 2003, a copy of which is available on the file of the present application. Briefly, executing a <i>source</i> command on the master JVM process 33 allows the warmup state, that is, the prewarmed state 37 and code cache 38, to be incrementally initialized. The application manager 32 sends the master warmup program 44 to the master JVM process 33 using the <i>source</i> command, which is executed to refine the context of the master JVM process 33. Each subsequently spawned cloned JVM process 33 inherits the then-warmed up context as inherited prewarmed state 43 and code cache 38 executed by the master JVM process 33 up to that time.

Master JVM Process Mapping

0029FIGURE 3 is a block diagram 60 showing, by way of example, a master JVM process 33 mapped into memory 62. Generally, the context for an executing process includes a data space, user stack, kernel stack, and a user area that lists open files, current directory and supervisory permission settings. Other types of context can also be provided. The context is stored and managed in the memory 62 by the operating system. At device boot time, the operating system instantiates a representation of the executable master JVM process 33 into the memory 62, possibly in non-contiguous pages 64a-d, and records the allocation of the memory space as page table entries 63 into the page table 61 prior to commencing execution of the master JVM process 33. In addition, the master JVM process context could similarly be mapped using other memory management systems, such as using demand paging, swapping and similar process memory allocation schemes compatible with process cloning, particularly process cloning with copy-on-write semantics.

Cloned JVM Process Mapping

0030FIGURE 4 is a block diagram 70 showing, by way of example, a master JVM process 33 and a cloned JVM process 34 mapped into memory 62 through memory space cloning. In a system with process cloning that does not provide copy-on-write semantics, physical copies of the pages 64a-c in the memory 62 storing the parent process context are created for each child process. In response to a process cloning request, the operating system instantiates a copy of the representation of the executable master JVM process 33 for the cloned JVM process 34 into the memory 62, possibly in non-contiguous pages 72a-d, and records the allocation of the memory space as page table entries 71 into the page table 61 prior to commencing execution of the cloned JVM process 34. Thus, the cloned JVM process 34 is created with a physical copy of the context of the master JVM process 33. Since a new, separate physical copy of the master JVM process context is created, the cloned JVM process 34 inherits the prewarmed state 37, including the preloaded classes of the master JVM process 33. However, the overall memory footprint of the runtime environment 31 is increased by the memory space required to store the additional copy of the master JVM process context.

Cloned JVM Process Mapping with Copy-On-Write

0031FIGURES 5A-B are block diagrams 80, 90 showing, by way of example, a master JVM process 33 and a cloned JVM process 34 mapped into memory 62 through memory space cloning with copy-on-write semantics. In a system with process cloning that provides copy-on-write semantics, only copies of the references, typically page table entries, to the memory space storing the parent process context are created for each child process. Referring first to FIGURE 5A, in response to a process cloning request, the operating system copies only the page table entries 63 referencing the memory space of the executable master JVM process 33 as a new set of page table entries 81 for the cloned JVM process 34. Thus, the cloned JVM process 34 uses the same references to the possibly non-contiguous pages 64a-d storing the master JVM process context as the master JVM process 34. Initialization and execution of the application associated with the cloned JVM process 34 requires less time, as only the page table entries 62 are copied to clone the master JVM process context. Furthermore, until the cloned JVM process 34 attempts to modify the master JVM process context, the memory space is treated as read only data, which can be shared by other processes.

0032Referring next to FIGURE 5B, the cloned JVM process 34 has attempted to modify one of the pages 82c in the memory space of the master JVM process context. In response, the operating system creates a physical copy of the to-be-modified memory space page 82c as a new page 91 and updates the allocation in the page table entries 81 for the cloned JVM process 34. Through copy-on-write semantics, the overall footprint of the runtime environment 31 is maintained as small as possible and only grows until, and if, each cloned JVM process 34 actually requires additional memory space for application-specific context.

Method for Dynamically Tracking Profiling Data Incrementally

0033FIGURE 6 is a flow diagram, showing a method 100 for dynamically and persistently tracking incremental profiling data in a process cloning application environment, in accordance with one embodiment of the invention. The method 100 is described as a sequence of process operations or steps, which can be executed, for instance, by the runtime environment 31 of FIGURE 2 or other components.

0034Initially, the application manager 32 is loaded (block 101). The master JVM process 33 is loaded and initialized at device boot time (block 102), as further described below with reference to FIGURE 7. Following loading and initialization, the master JVM process 33 enters an inactive sleep mode (block 103). Upon receiving a request from the application manager 32 (block 104), the master JVM process 33 awakens (block 105). If necessary, the master JVM process 33 checks the network connection identifier (ID) (block 106) for the application manager 32 and determines the type of request (block 107). The master JVM process 33 recognizes the commands <i>list, jexec, source,</i> and <i>kill,</i> as described above with reference to FIGURE 2. If the request type corresponds to a <i>source</i> command, instructing the master JVM process 33 to execute the master warmup program 44 (block 108), the master JVM 33 executes the master warmup program 44 (block 109), which refines or "enriches" the prewarmed state of the master JVM process 33, such that subsequently cloned JVM processes 34 spawned from the master JVM process 33 will inherit the refined context. If the request type corresponds to a <i>jexec</i> command, instructing the master JVM process 33 to initiate an execution of an application through process cloning (block 110), a cloned JVM process 34 is cloned and executed (block 111), as further described below with reference to FIGURES 8 and 9.

0035Concurrently, the application manager 32 collects the profile data from the memory buffer 52 or stored profiles 48 (block 112). Each profile collector 39 collects and stores incremental profile data generated by a profiler 40 executing in one or more cloned JVM processes 34. The dynamically profiled data is polled regularly by the application manager 32 to create the master dynamic profile 49, which is fed back into the master JVM process 33 using the <i>source</i> command (block 108) to benefit subsequent cloned JVM processes 34. Thus, if the master JVM process 33 is to be updated through feedback of incremental profile data (block 113), the application manager 32 warms up the master JVM process state by sending a <i>source</i> command (block 114). The master dynamic profile 49 also forms the basis of a persistent profile that can be used by the master JVM process 33 upon the next device boot. If maintained persistently (block 115), the master dynamic profile 49 is stored persistently, such as in a file (block 116). Processing continues indefinitely until the application manager 32, master JVM process 33, and the runtime environment 31 are terminated.

Routine for Loading Master JVM Process

0036FIGURE 7 is a flow diagram showing the routine 120 for loading a master JVM process 33 for use in the method 100 of FIGURE 6. One purpose of the routine is to invoke the master JVM process 33 and to warmup the master JVM process state for inheritance by cloned JVM processes 34.

0037Initially, the master JVM process 33 begins execution at device boot time (block 121). The master JVM process 33 executes two sets of warmup actions using the master warmup program 44 and the master dynamic profile 49. The master warmup program 44 is static and does not generally change between device boots, whereas the master dynamic profile 49 is subject to continual change based on the incremental profile data collected and assimilated by the application manager 32. First, the master JVM process 33 executes the master warmup program 44 to form a memory image that resembles that of an initialized, primed and warmed up JVM process with key classes, methods and initialized state stored in the master JVM process context (block 122). Second, if the application manager 32 maintains the master dynamic profile 49 persistently (block 123), the master JVM process 33 retrieves and executes the master dynamic profile 49 (block 124). The master JVM process 33 then completes any other warmup operations (block 125) and the routine returns.

Routine for Process Cloning without Copy-On-Write

0038FIGURE 8 is a flow diagram showing the routine 130 for cloning a process without copy-on-write for use in the method 100 of FIGURE 6. One purpose of the routine is to create and initiate execution of a cloned JVM process 34 through process cloning that does not provide copy-on-write semantics.

0039Initially, the memory space containing the context of the master JVM process 33 is physically copied into a new memory space for the cloned JVM process 34 (block 131). Optionally, the master JVM process 33 can set operating system level resource management parameters over the cloned JVM process 34 (block 132), including setting scheduling priorities and limiting processor and memory consumption. Other types of resource management controls are possible. The cloned JVM process 34 is then executed by the runtime environment 31 (block 133) using the duplicated master JVM process context. During execution, the cloned JVM process 34 profiles the application to create incremental profile data (block 134). If the incremental profile data is stored persistently (block 135), the cloned JVM process 34 stores the incremental profile data as stored profiles 48 (block 136). Otherwise, the incremental profile data is transiently staged in the memory buffer 52 (block 137). The routine returns upon the completion (block 138) of the cloned JVM process 34.

Routine for Process Cloning with Copy-On-Write

0040FIGURE 9 is a flow diagram showing the routine 140 for cloning a process with copy-on-write for use in the method 100 of FIGURE 6. One purpose of the routine is to create and initiate execution of a cloned JVM process 34 through process cloning that provides copy-on-write semantics.

0041Initially, references to the memory space containing the context of the master JVM process 33 are copied for the cloned JVM process 34 (block 141). Optionally, the master JVM process 33 can set operating system level resource management parameters over the cloned JVM process 34 (block 142), including setting scheduling priorities and limiting processor and memory consumption. Other types of resource management controls are possible. The cloned JVM process 34 is then executed by the runtime environment 31 (block 143) using the referenced master JVM process context. During execution, the cloned JVM process 34 profiles the application to create incremental profile data (block 144). If the incremental profile data is stored persistently (block 145), the cloned JVM process 34 stores the incremental profile data as stored profiles 48 (block 146). Otherwise, the incremental profile data is transiently staged in the memory buffer 52 (block 147). Each time the cloned JVM process 34 attempts to write into the memory space referenced to the master JVM process context (block 148), the operating system copies the applicable memory segment (block 149). Otherwise, the cloned JVM process 34 continues to use the referenced master JVM process context (block 150), which is treated as read only data. The routine returns upon the completion (block 151) of the cloned JVM process 34.

0042The data structures and code described in this detailed description are typically stored on a computer-readable storage medium, which may be any device or medium that can store code and/or data for use by a computer system. This includes, but is not limited to, magnetic, semiconductor, and optical storage devices such as disk drives, magnetic tape, flash memory, CDs (compact discs) and DVDs (digital versatile discs or digital video discs), and computer instruction signals embodied in a transmission medium (with or without a carrier wave upon which the signals are modulated). For example, the transmission medium may include a communications network, such as the Internet.

0043The foregoing descriptions of embodiments of the present invention have been presented for purposes of illustration in order to enable a person skilled in the art to appreciate and implement the invention. They are provided in the context of particular applications and their requirements, but are not intended to be exhaustive or to limit the present invention to the forms disclosed. Accordingly, many modifications and variations will be apparent to practitioners skilled in the art, and the scope of the present invention is defined by the appended claims and their equivalents.