EP1145109A2 - Procede et systeme d'etablissement et d'utilisation de la puissance de traitement informatique reseautee en veille - Google Patents

Procede et systeme d'etablissement et d'utilisation de la puissance de traitement informatique reseautee en veille

Info

Publication number
EP1145109A2
EP1145109A2 EP00959899A EP00959899A EP1145109A2 EP 1145109 A2 EP1145109 A2 EP 1145109A2 EP 00959899 A EP00959899 A EP 00959899A EP 00959899 A EP00959899 A EP 00959899A EP 1145109 A2 EP1145109 A2 EP 1145109A2
Authority
EP
European Patent Office
Prior art keywords
client
computer
provider
computational power
tasks
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
EP00959899A
Other languages
German (de)
English (en)
Other versions
EP1145109A3 (fr
Inventor
Steven L. Armentrout
James O'connor
James Gannon
Brian Sletten
Sean Cier
Sarah Carlson
Jonathan Davis
Greg Dupertuis
Scott Mccloughlin
Antony Davies
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Parabon Computation
Original Assignee
Parabon Computation
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Parabon Computation filed Critical Parabon Computation
Publication of EP1145109A2 publication Critical patent/EP1145109A2/fr
Publication of EP1145109A3 publication Critical patent/EP1145109A3/fr
Withdrawn legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5061—Partitioning or combining of resources
    • G06F9/5072—Grid computing

Definitions

  • This invention relates generally to computer use over a network to form a distributed computing platform. More specifically the present invention is a system and method for use of idle time of computers that are connected to a network by users requiring significant computing power without the need for a large scale dedicated processing unit.
  • the present invention is drawn to organization and use of the idle processing power of general-purpose computers to form a distributed computing platform. It is therefore an objective of the present invention to provide large amounts of computational power to users without the users having to purchase a large computer for such purposes. It is a further objective of the present invention to harness the idle computational power of many computers and make that power available to clients on an as needed basis. It is yet another objective of the present invention to allow those who provide computers to be used to have complete access to their computers whenever desired without interruption of personal use. It is still another objective of the present invention to provide computational power without regard to any specific schedule of computer non-use. It is yet another objective of the present invention to allow clients who require computational power to specify the characteristics of the power they require.
  • the centralized task server comprises a three-tiered architecture as more fully set forth below and is redundant, fault tolerant and scalable as traffic and clients increase.
  • a task scheduler in the CTS matches power requested to available provider computer resources.
  • the client is the individual or organization that is requesting the data processing services of the present invention.
  • the provider constitutes the plurality of individuals or organizations having excess processor capacity that are offering that capacity to the system of the present invention to process requests from clients.
  • the flow of the present invention is as follows: 1. Using the client software from their desktop, clients launch their distributed programs, the tasks of which are transported to the centralized task server. 2. Client tasks are pooled at the centralized task server and allocated to provider computers based on a scheduling algorithm that takes into account the characteristics of the provider computer (e.g., processor speed, disk space, amount of RAM, communication speed, percentage of time the provider computer is on-line; percentage of time the provider computer is in use).
  • the CE on the provider computer periodically contacts the server of the present invention to retrieve task assignments or other control messages. Such contact occurs periodically and opportunistically (whenever a network connection is present, for example, when a provider computer has dialed into the Internet) and not just when idle computational power is detected.
  • the invention also allows for an auto-dial capability where the provider computer can specify intervals when the CE can connect to the server. Multiple tasks can be assigned to a CE at a given time. This is done to increase the probability that the CE will always have tasks to execute. 3.
  • a provider computer's CE detects that its host computer is idle, it executes tasks that were retrieved from the centralized task server. 4. The results of completed tasks are saved locally.
  • the present invention represents a gigantic virtual computer or distributed computing platform, ideally suited for performing large-scale parallel distributed computation, a function that was formerly the exclusive domain of traditional "chips-and-solder" supercomputers. The present invention, however, will perform such computations at a fraction of the cost and on an as-needed basis.
  • the business model associated with the present invention is to create a new market for idle computation power.
  • a business running the apparatus and method of the present invention will purchase the idle computational power of millions of providers (individual computer owners), and deliver that idle computational power as a single entity to clients (technologists seeking ultra-high-performance computation).
  • idle computation has no value whatsoever because the market for computation is inseparable from the market for computers. Users have no means of buying, for example, larger amounts of times on disparate computers to execute a job when the user needs it and not just when a specific computer is available.
  • immense value can be created by a computational intermediary in the same way that financial engineers create value by constructing derivative financial instruments.
  • the present invention strips excess computational capacity from provider computers, bundles it into quantities suitable for large- scale parallel computations, and delivers it to clients on an as needed basis.
  • Providers benefit by selling a heretofore-wasted resource.
  • Clients benefit because they can inexpensively purchase just the quantity of computation needed, thereby avoiding the normally huge fixed costs associated with purchasing a supercomputer.
  • the business model of the present invention creates and intermediates the new market in idle computation.
  • an applications research and development group will add value by utilizing its superior knowledge of distributed processing to solve high-payoff computational problems. Such a team will spur demand in two ways: First, its purchases of idle computation will stimulate market activity; second, the success of applications and research group will demonstrate to prospective clients the power and cost-effectiveness of idle time processing.
  • Figure 1 illustrates an architecture of one embodiment of the present invention
  • Figure 2 illustrates a flow of work over one embodiment of the present invention.
  • Figure 3 illustrates one embodiment of the CE architecture of the present invention.
  • Figures 4A and 4B illustrate a normal execution scenario for tasks.
  • Figure 5 illustrates an overall architecture of a server of the present invention.
  • Figure 6 illustrates a determination of processing time of virtual nodes.
  • Figure 7 illustrates communications paths of the present invention.
  • Figure 8 illustrates expected time for completion of a task probability.
  • the present invention is a system and method for allowing those people who own personal computers or larger computers, (hereinafter "providers") to offer unused and excess computational power on their “provider computers” to people who have need for large computational power for their computing tasks. These people are known as “clients.”
  • clients These people are known as “clients.”
  • the present invention basically comprises three major components: a client application that has been developed within a client development environment; a centralized task server (CTS) 26; and a compute engine (CE).
  • CTS centralized task server
  • CE compute engine
  • a client application is an application program, developed within the client development environment that is part of this invention, which solves a problem utilizing the system that constitutes this invention.
  • the typical problem such an application solves is an extremely computation intensive problem that is amenable to decomposition into a large number of independent tasks (this is not intended as a limitation).
  • the client development environment consists of a runtime API which defines how tasks interact with their environment, a client API which allows a client application to manage the execution of jobs consisting of a very large number of tasks, and a set of tools that support the monitoring and control of very large jobs.
  • a centralized task server (CTS) 26 intermediates the needs of client applications and the power of the provider computers 30-36 to meet the needs of clients for computational power.
  • the CTS receives tasks from clients and assigns them to one or more providers based the characteristics of the available providers and the job characteristics specified by the client. When providers return results, the CTS forwards these back to the client application.
  • Each provider computer has a resident compute engine (CE) that receives and executes the various tasks that are assigned to it by the CTS.
  • CE resident compute engine
  • the CE runs client tasks and manages the use of idle computation power of the provider computer.
  • the CE returns intermediate and final results to the CTS that then forwards them back to the client application.
  • Communication between the client application, CTS, and CE is governed by a communication protocol unique to the present invention.
  • the message channel is a bi-directional channel between clients and the server that is used to pass high-level messages that facilitate distributed processing.
  • the data channel is a channel used to pass large blocks of data called data elements and executable elements.
  • Executable elements are blocks code developed by clients or a third-party. This code represents tasks or portions of tasks.
  • Data elements are blocks of data that are either input to or output of client tasks.
  • High-level messages transmitted on the message channel contain references to data and executable elements.
  • Clients 10 that is, those entities desiring execution of computationally intensive tasks, launch client software that submits jobs consisting of multiple tasks. Although only one box is shown, it is anticipated that there will be many clients that have such computationally intensive jobs on or about the same time. Thus more than one client will be requesting such services.
  • the centralized task server (CTS) takes all incoming tasks and pools them in some priority order.
  • the CTS validates each task (checks that all required executable and data elements are in place) and assigns it to one or more CEs, based on information on available provider computers and their characteristics.
  • the software of the CTS will be able to size the available computational power to the tasks that are pooled thus assigning the tasks the appropriate amount of available computational power.
  • the CE contacts the CTS to see if there are any messages for it. These messages could be task assignments or other control messages.
  • the CE retrieves tasks from the CTS so that, to the degree possible, the tasks are readily available for execution whenever idle computing power is available.
  • a CE receives a task assignment from the CTS, it downloads the data and executable elements required to run that task.
  • the task is then put in the "prepared” state.
  • Each CE constantly monitors its processor state and when the processor is idle it launches the "next" prepared task.
  • the task selected to run is a function of the arrival time of the task, the task state, and the task priority. If a task is received that has a higher priority than the currently running task, the current task is preempted (halted and results saved) and the new task is executed. Once tasks are executed, task results are returned 10 to the CTS.
  • the CTS journals all work done by provider computer CE's on behalf of clients. This journal is then used as the basis for adjustments to client and provider computer accounts on a periodic basis using known accounting methods. These accounts are the basis for client billing and provider payment. As mentioned above, clients develop applications using the client development environment that is part of this invention.
  • the client development environment consists of a runtime API which defines how tasks interact with their environment, a client API which allows a client application to manage the execution of jobs consisting of a very large number of tasks, and a set of tools that support the monitoring and control of very large jobs. These are described in more detail below.
  • the runtime API defines the environment in which client task run.
  • the runtime API consists of a set of function calls that the client task will exercise during execution to interact with its environment. This API contains calls that the task can use to: • Obtain access to the initialization parameters for the task. • Obtain access to data elements needed to perform its function. • Perform a checkpoint so that if the task is interrupted, it can restart at the checkpoint and therefore reduce the loss of work. • Send interim and final results back to the server.
  • the API defines entry points that allow the environment to control the task. These include: • Start a task (note that restarting a task is a special case of starting a task). • Stop (suspend) a task. • Request a checkpoint. • Request a status. • Terminate a task.
  • the two most important implementations for the purpose of this invention are the implementation in the CE and the run local implementation included in the client development environment.
  • the CE implementation allows the task to run in the CE on a provider's box (regardless of platform).
  • the run local implementation allows clients to execute tasks on their local client machine.
  • the client API is used to develop an application to control the execution of a job running on the present invention.
  • the client API broadly supports job definition, job launch, job monitoring, and job control.
  • Job definition involves the creation of the job, setting job properties (e.g., requested power, deadline, maximum cost, scheduling hints), creating its data/executable elements and its tasks, and associating data/executable elements with tasks.
  • job properties e.g., requested power, deadline, maximum cost, scheduling hints
  • the client has the option of providing certain limits to the job being submitted. For example, a client can limit the job to a certain dollar expenditure or amount power applied.
  • Job launch involves launching the tasks that constitute a job.
  • Job monitoring involves the selective monitoring of a job.
  • the data available included both status and statistics.
  • the client API allows the client to attach to a job or a specific task and monitor all or a subset of the results being returned by the job or task. Tasks can return interim and final results via the runtime API. Certain default patterns are set by the system concerning when interim results are to be obtained although the client can change these parameters as well as explicitly request results for executing jobs.
  • Job control functions include starting, removing, stopping, and resuming jobs and tasks.
  • the client API of the invention supports jobs of arbitrary size. A "fire and forget" event model allows clients to launch jobs much larger than the client machine would otherwise be capable of tracking simultaneously. This is accomplished by programmatically ensuring that clients monitor only task results, not the tasks themselves.
  • the invention also includes general-purpose tools, developed using the Client API, for monitoring jobs. Jobs can be monitored whether they run locally (at the client's node) or remotely (on a provider computer node).
  • the client display is updated periodically (the periodicity is configurable).
  • the display for the client is designed to keep the client fully informed at all times regarding the job submitted. As such it includes dollars spent, number of tasks running, number of completed tasks, number of task that are executing, the number of task delivered to the CTS, the total work accomplished, the amount of work done per task (percentage), the work done on completed tasks, and the time spent per completed task. These statistics are illustrative only.
  • the CE is the software unit that resides on a provider computer and accomplishes the timely and secure execution of client tasks.
  • the CE 84 manages the interface between the provider computer and the server of the present invention, monitors the state of the provider computer's machine, schedules the execution of tasks on the provider computer, and manages all communication with the server of the present invention.
  • the CE 84 receives tasks from the CTS, posts results to the server, provides status of tasks and receives messages to remove, suspend or resume tasks.
  • the CE 84 can be configured to execute tasks continuously, albeit at low priority (so as not to interfere with a provider's normal use of their computer), or to start execution of tasks when a provider's computer is idle (i.e., not being interactively used) and stop execution immediately upon provider interaction with their computer. In this way, the CE is designed to be unobtrusive to the provider.
  • CE 84 communications with the server and execution of tasks occur independently.
  • the CE communicates with the server, provided a network connection is available (e.g., when a provider has dialed into the Internet). In so doing, the CE receives elements that are placed in the task queue 86 to be executed whenever possible. Thus, obtaining and staging of client tasks takes place at a time independent from the execution of those tasks. Making CE communications independent from CE task execution adds to the efficiency of the overall system.
  • the CE also includes a dialing scheduler that can be preset by the provider to dial out to the server of the present invention at configurable intervals and only during preset periods. Thus, for example, the provider can direct the dialer to dial out to the server every ten minutes, but only between the hours of 2 a.m.
  • the entire CE resides on the provider computer, which as noted earlier, can be any form of computer having excess capacity.
  • provider computers that run the Windows® operating system, the Macintosh® operating system, the Linux® operating systems and others are all candidates for use with the present invention.
  • Various security measures are taken in executing tasks using the present invention.
  • the CE uses a virtual machine 78 to execute client tasks 82. Examples of virtual machines, but not intended as a limitation, include the Java Virtual Machine and the C# Virtual Machine. Throughout most of the remainder of the discussion we refer to the CE virtual machine as the Java Virtual Machine or JVM 78. Security restrictions on the virtual machine prevent client tasks from accessing resources on the provider machine directly.
  • the virtual machine is altered such that only encrypted tasks provided by the server can be executed. This both protects tasks when stored on disk and ensures that the CE executes only valid tasks. Further, the system has the option of encrypting everything that is written to disk both in the provider computer and on the server.
  • FIGs 4A and 4B a normal execution scenario for tasks is illustrated. It is important to note that the CE 84 a highly capable application. It has the ability to run tasks constantly at low priority so that a provider's normal use of the computer is unimpeded; alternatively, it can actuate in much the same way that a screen saver actuates, executing tasks only when a provider is not using their computer.
  • the CE will exercise a small window that will be visually perceptible by the user of a particular computer of the present invention.
  • the CE window will "pop up" with the cursor within the CE window.
  • the application window can also simply be illustrated with an actual Screensaver, but the description below assumes the CE window is an application window.
  • the CE window activity is independent from the execution of tasks, but can be synchronized with such so as to provide, among other things, status information about the execution of the current task.
  • Various buttons and areas within this CE window will be accessible to the user such that accounts can be accessed, amount of time associated with use of the users computer can be determined.
  • the graphic window of the CE is displayed for the provider when the CE detects the host computer is idle. Its presence indicates to the provider that the system is processing, or is ready to process tasks in the queue, and has a number of functions.
  • the window can show the state of processing, that is whether tasks are being processed at a particular moment or not. Further, within the window are located various other buttons that may be clicked to determine the status of the provider's account. In addition, advertising or other messages may be presented in the CE window, which will allow a user to proceed to a particular website to view further information concerning the messages.
  • the CE window can operate in the nature of a browser. The size and placement of the CE window during "pop up" is configurable by the user.
  • the system seizes control of the cursor and locates it at any configurable spot within the CE window.
  • the cursor can be placed in a spot where messages are to be displayed or in a location to actuate any functionality that is available to the provider while the CE is operating, such as the quick disappearance of the display window as described above.
  • the user can also confine the CE to a system tray if desired.
  • the icon or panel in the tray can be used to indicate, among other things, engine status, the arrival of new display content or upgrade packages, or Internet connectivity status.
  • the CE can be configured and executed as a "daemon" having only a command-line user interface.
  • the CE also supports the dynamic display of marketing content in the CE window to the provider computer.
  • the server notifies the CE when new content is available and where it can be downloaded.
  • the user will simply type a key or cause the cursor to be moved outside the CE window.
  • the CE will shut down all of the processing of the present invention in an orderly fashion.
  • the CE can be configured to execute tasks constantly, albeit at low priority, in which case the CE window can be made to disappear as described above, yet tasks will continue execution at lower priority. If the user would like to interact with the window, then a single click anywhere on the window will cause it to enter a new state where it will not automatically disappear. This contributes to the ease of use of the CE.
  • This state is exited when the user closes the window or after a period of inactivity.
  • a CE there are a series of execution steps when a CE is running on a user's computer.
  • the CE obtains a task message from the server 90.
  • the executable elements from the element server are written to the disk of the computer.
  • the disk management logic within the CE ensures that any such storage will not cause the CE to exceed its storage space budget 92.
  • the CE then writes out the task policy file that establishes the permissions for the Runtime 80, and the task to be run 82. This task policy ensures that the client task can neither access the provider computer's disk or the network.
  • the runtime policy gives the runtime module 80 access to data elements need to run the task (if any) and authorizes the runtime module 80 to connect back to the Core Engine.
  • the CE then launches the virtual machine (JVM) 96 and executes the Runtime 80.
  • the path of the policy file, the security manager, the maximum size to be used based upon the CE's memory budget, and the class path including all executable elements, are passed to the JVM 96.
  • the class name for the task and the CE communications port are also passed as parameters to the runtime module 80.
  • the JVM then launches the Runtime main module 98. This runs in a protection domain that allows the Runtime to open a network connection to the CE and to read executable and data elements from the disk.
  • the JVM opens a server socket, connects to the CE port passed in on the command line, and sends a message containing its server port 100.
  • the CE then connects to the JVM's server port. This results in two connections between the CE and the JVM, i.e., a CE to runtime channel, and a runtime to CE channel 102.
  • the runtime requests the parameters for the task over the runtime to CE channel 104 and the CE returns those parameters in response over the CE to runtime channel.
  • the runtime requests the data elements for the task 106 over the runtime to CE channel, and the CE returns them in response over the CE to runtime channel.
  • the runtime then establishes a task context that the task uses to communicate with the runtime system 108.
  • the runtime establishes and begins the client task 110.
  • the executable elements for this task having been provided to the runtime via the class path provided for the JVM when the JVM was started.
  • the client task then runs in a separate protection domain that prevents it from accessing the network or the disk. This protection domain is established through the permissions granted to the task in the task policy file.
  • the client task interacts with the runtime through the task context object established for it by the runtime 112.
  • the runtime calls run with the client tasks privileges when possible. If the runtime needs to perform an operation requiring additional privileges, i.e., accessing a particular data element, the runtime executes a privilege block that takes on the minimum privilege for the minimum amount of time required to complete the operation.
  • the client interacts with the runtime to get task parameters and to access task elements 114.
  • the client task can send out temporary results using the set status call to the runtime 116, these results are passed to the CE over the runtime to CE channel.
  • the CE then sends these interim results in the form of a task status to the server.
  • the client task can create a task checkpoint using the checkpoint call to the runtime 118, the checkpoint is passed to the CE over the runtime to CE channel.
  • the CE serializes the checkpoint to disk so it is available in case a task restart is required.
  • the client task does a set status runtime call 120 with the final results and exits. The results are passed through the CE over the runtime to CE channel.
  • the CE then sends these results in the form of a task status message to the server. After the CE sends a final task status noting that the task is complete 120, the space that is consumed by the task definition and related data can be reclaimed for use in running other tasks. Once the task completes 122, the runtime closes down its connections and exits. As noted above, the CE is a very flexible program that does not impair the ordinary use of the computer in any way. For example, when the user moves the mouse cursor outside of the CE window, the CE sends a shut down request to the Runtime. If the JVM does not shut down in seconds, the JVM is automatically stopped. When the Runtime receives a shut down request, it calls the clients task stop method.
  • the client is allowed to send a set status and create a checkpoint before exiting. This reduces the loss of work and allows the program to restart at the checkpoint when processing resumes later. If the client task supports checkpointing, the runtime sends the checkpoint back to the CE on the runtime to CE channel. If a checkpoint is received, the CE writes it to disk. Once the client task exits, the runtime closes its connections and exits.
  • the CE has been configured to execute tasks intermittently based on user interaction (versus continuously), then when the CE detects that the PC has been idle for a configurable period of time, and thus is available to run a task, it checks its queue for a paused task, and restarts the task from the last checkpoint which was previously stored on disk, using the same mechanism used to start the task at the outset. If there was no paused task, the next task in the queue is run.
  • the CE functionality while used in the present invention, could also be used in other fashions. This would, for example, allow particular programs to be run within a given office, thereby utilizing the full power that is available on all PCs within an office location.
  • the centralized task server intermediates tasks submitted by a client application time and the CEs on providers' computers that have excess computational capacity to run those tasks.
  • the CTS registers clients, accepts tasks for execution, distributes those tasks to computers, receives results back from the provider computers, conveys results to the client computers, charges clients for the computational work performed and pays providers for the use of their computers.
  • One of the most critical functions of the CTS is task scheduling. Part of the scheduling process is to provide clients of the present invention, who desire to obtain the aggregate computational power, with the desired amount of processing power on demand.
  • the server of the present invention loads tasks onto a single provider computer based in such a way as to match computer capability with task requirements. As tasks are completed, and through the communications channel between the provider computer and the server, additional tasks are fed to the provider computer for execution. All new tasks are placed in a task execution queue on the provider's computer for subsequent execution.
  • the CE supports the concept of task priorities and task preemption (tasks can preempt executing tasks of lower priority). These features allow the server to more effectively utilize available provider computer resources.
  • the server of the present invention In order for the server of the present invention to correctly characterize the provider's computer, the server also collects usage and performance histories (profiles) from the provider computer concerning a wide variety of parameters.
  • the server collects Internet connectivity patterns and processor availability patterns during the course of any 24-hour period, weekly period, monthly period, etc.
  • the processing power of the processor itself is also important since certain clients may request higher or lower processing power depending on relative cost and timeliness tradeoffs.
  • the task scheduler must fulfill the following requirements: The task scheduler gets jobs and associated tasks from clients. It must then take those tasks and distribute them to various provider computers for execution. It must also record the fact that it has assigned a particular task to a particular provider computer. The task scheduler passes both interim and final results to the client and records the amount of work performed to an on-line ledger. It then credits a provider's account with the amount of time that the provider computer actually performed on the specific task.
  • the task scheduler also logs completed work in a client's account noting the amount of work performed so that an appropriate billing can be passed on to the general ledger module for billing to the client. Cancellation of tasks or jobs (collections of tasks) from a client is also processed by the task scheduler. It accepts any job cancellation from the client and insures that appropriate communication is sent throughout the system of the present invention canceling a particular job.
  • the task scheduler automatically provides e-mail or other automated notification to the client that the job is complete.
  • Results that are received by the task scheduler are archived if the client has not retrieved those results. From time to time the task scheduler deletes unclaimed information from the data archive. Since clients pay for the computational power they use, the task scheduler periodically checks the on-line ledger to determine if a client has exceeded an account limit. If so, the task scheduler notifies the client and, if necessary, cancels all jobs for the client until such time as the account is brought current. In the event that a particular provider computer that is working on a particular task does not respond for a pre-set period of time, the task scheduler moves the task to another provider computer. Thus, it must be able to detect unresponsive CEs on provider computers.
  • the task scheduler also can elect, through its software, to launch redundant tasks to different provider computers to insure timely completion of any particular task.
  • the CTS is also tasked with delivering the appropriate amount of compute power for a client task, not more or less, for a given client. Since charges for the client are made in part based upon the compute power delivered, this assembly of provider computers and delivery of computational power must be accurate.
  • the task scheduler also balances data processing loads among the various provider computers based upon the capabilities of the provider computer and the bandwidth that is available to provide elements and tasks to provider computers.
  • the CTS collects statistics regarding provider computer capabilities. This is critical for both assigning payment to a provider and a debit notice to the client.
  • the CTS may send the same tasks to more than one provider computer. This helps the CTS be assured that the computations that are performed by the assigned provider computers are correct. Since results from provider computers should be the same for the same task (unless randomization is integral to the calculation of the task), any differences in response are noted by the CTS and resolved. This assures the client that answers are accurate.
  • the CTS takes explicit measures to protect the intellectual property inherent in a task. To accomplish this, the CTS obfuscates executable elements and data elements to remove contextual clues to the code's intended purpose. Furthermore, the CTS also removes from tasks all traces of client identity.
  • a provider has no means of knowing the clients for whom elements are executed nor their domain of inquiry.
  • the CTS executes certain benchmark tasks on provider computers.
  • various test tasks with known results are sent to the provider computer to ensure that results returned by the provider computer are accurate.
  • Unexpected results from a provider can be used to identify attempts to submit fraudulent results or otherwise tamper with the normal execution of the system of the present invention.
  • the present invention allows the client to specify job and task parameters via the client interface. Using the client interface, the Client selects desired attributes of nodes in a virtual cluster.
  • the client also provides information on his job: estimated work per task (in GF-Hours), incoming data (in MB), outgoing data (in MB).
  • the CTS may construct virtual nodes (one or more nodes operating on the same task so as to improve the expected time-to- completion of a given task). Virtual nodes are discussed in more detail elsewhere in the document.
  • the client After the client defines the tasks to be executed, they are provided to the central server of the present invention for "launching,” i.e., the placement of the task in a queue for scheduling and ultimate distribution to provider computers. For billing and compensation purposes, a task is "launched” when it is scheduled. In this way, the latency from a task sitting in queue is attributed to the provider computer and so considered part of the provider computer's intermittent "unavailability".
  • the system of the present invention records the time at which each task is launched to the first provider computer in a virtual node. The system also records the time at which the first completed task is returned from a virtual node. At any point in time, the amount of power running through the virtual cluster is the sum of the powers of the virtual nodes running.
  • the system constantly monitors virtual nodes for possible failure. Failure is defined as an absence of communication from a configurable number of provider computers in a virtual node for a configurable amount of time. When a virtual node fails, the task assigned to it is re-launched on another (perhaps faster) virtual node.
  • the client can be billed.
  • the client charge is the price per hour (based on power running through the cluster) for each hour (or portion thereof) since the last billing. For example, and referring to figure 6, the determination of running time on providers' computers is illustrated.
  • each virtual node in the cluster operates at 0.1 GF.
  • the client is charged an hourly rate for running tasks based upon the power, P, running through the client's account at each 15-minute increment.
  • the client is charged an additional amount per GB for data transferred through each of the four communications paths, 150, 152, 154, and 156 of the present invention as illustrated in Figure 7.
  • the communications paths 150, 152, and 154 represent duplicated data and are a function of the degree of reliability the client achieves. Of the dashed arrows, only one will be executed, and so the client will be charged for only one of these paths.
  • the client is also charged for data transfer during the course of executing the particular tasks of the client. For example, let d* be the quantity (in GB) of data transferred from the client to the server. Let d 0 be the quantity of data transferred from the server to the client. Given that some of the data will be duplicated for processing by virtual nodes comprised of more than one provider computer, the client's total data transfer charge is given by the following:
  • g is the transfer cost per leg per GB
  • r is the requested node reliability
  • ⁇ is the average provider computer reliability.
  • Requested node reliability r can be defined for a virtual node of CPU speed C and bandwidth B as the probability of the virtual node completing a task in no more time than the time it would take a fully idle, constantly connected computer with CPU speed C and bandwidth B to complete the task.
  • Average provider computer reliability ⁇ is defined as the probability of the provider's frequency of contact not diminishing, in the near term, from the provider's historic average.
  • the client is also charged an hourly premium for requesting reliability greater than a set baseline reliability. The charge per hour is: Price per hour for
  • k is the baseline price per MHz-hour (provider computer price per hour for the ideal machine)
  • C is the CPU speed requested of the virtual nodes
  • N is the number of virtual nodes running
  • r is the requested reliability
  • R is the baseline reliability
  • a CPU availability rating, a c is constructed along with a bandwidth availability rating, a t ,. These ratings measure the average availability of the provider computers' CPU and bandwidth over time. Unlike in the full version of the task scheduling model (described below), these ratings are single values versus vectors.
  • the system of the present invention records transitions of CPU and bandwidth (between "available” and "unavailable") over time. Defining each minute of availability as having value 1 and each minute of unavailability as having value 0, the system takes the moving average of these values over the past seven days to obtain average availabilities for CPU and bandwidth.
  • the system For each provider computer, the system next measures CPU speed, C, and bandwidth, B (through benchmark tasks). These measures are periodically updated to ensure that the system has knowledge of the full availability of the provider computer's system.
  • candidate provider computers have expected bandwidths (bandwidth multiplied by bandwidth availability) that satisfy the following equation:
  • the system calculates the weighted average availability, ⁇ , as: w d t +d divide — + . ⁇ C B w d : + d condiment
  • the system arbitrarily groups the provider computers into virtual nodes such that the probability of the virtual node returning the task in, at most, T R equals r. Specifically, the system randomly picks provider computers from the subset and groups them into virtual nodes such that, for reliability r, and N nodes, each with weighted availability ⁇ ,,
  • N the value of N that satisfies the equation may be non-integer.
  • the server will launch a different number of duplications of each task such that the average number of duplications per task equals N.
  • the system launches the task simultaneously on all the provider computers that comprise a virtual node. As noted earlier the task is "launched” when it is scheduled. When the first provider computer returns the completed task or when the client cancels the task, all duplicate tasks on remaining provider computers are cancelled. We define "cancelled” as the placing of a "remove task" command in the queue. A particular provider computer's task is considered concluded when the first of the following occurs: • The provider computer receives notice that the task has been cancelled. • L hours pass since the task was cancelled.
  • the system For each provider computer, the system computes the time that elapsed between the launching of the task and the conclusion of the task.
  • the amount of work the provider computer completed is estimated as the time that elapsed multiplied by the provider computer's power rating. Note that, effectively, the provider is being paid for time, not work.
  • the provider computer's power rating is a function of the provider computer's availability
  • the rate of payment the provider receives is a function of the average work per unit time the provider computer completes over time.
  • the system established a target annual payment per MHz-hour, k, and a baseline reliability, R.
  • the system constructs a CPU signature and a bandwidth signature. These signatures show the probabilities of the provider computer's CPU and bandwidth being available for a fixed time interval over time. For example:
  • the number of intervals should be enough to cover a reasonable circadian "cycle” (e.g. 1 week).
  • the signature should be updated periodically.
  • the system For each provider computer, the system measures CPU speed by sending to the provider computer certain benchmark tasks that have known response times. The bandwidth is also measured. As noted earlier, these measures are periodically updated. The system then combines the signatures with CPU speed and bandwidth to construct a profile matrix.
  • the profile matrix shows the probabilities of the provider computer completing a task of given work, input data, output data, and time of launch within certain time lengths.
  • Example: The tree below shows all possible paths of success and failure of a provider computer working toward completion of a task. Each step down in the tree represents a time increment. Left branches represent failures to work. Right branches represent successes toward work. Probabilities are associated with each branch based on the CPU and bandwidth signatures.
  • the expected time-to-completion is the sum of the products of the probabilities of reaching each terminal node and the times required to reach the nodes. This concept is illustrated in Figure 8.
  • the client specifies the CPU speed and bandwidth of a virtual node.
  • the system finds all provider computers that have an expected time-to-completion equal to the time-to- completion for a fully available node with the specified CPU speed and bandwidth. Call this time-to-completion the "desired time-to-completion".
  • the system of the present invention arbitrarily groups these provider computers into virtual nodes such that the probability of the virtual node returning the task in the desired time-to-completion equals some fixed value.
  • the system launches the task simultaneously on all the provider computers comprising the virtual node.
  • the system computes the work each provider computer performed from the time the tasks were launched until the first task was returned. We calculate work as the number of hours the provider computer worked multiplied by the provider computer's power rating. Providers are paid a fixed amount per unit work. Payment is made on a per GHz-hour rate, T, and a target reliability r.
  • the server comprises a registration servlet 50, which accepts registration requests over the network 28 from clients and providers who wish to participate on the system. Registration information is sent by the registration servlet to the registration manager 52 for processing.
  • the registration manager 52 receives appropriate information from the registration servlet 50 and assigns the client/provider computer to a database. The registration manager then generates a certificate and returns the certificate in the response to the client/provider computer.
  • the provider computer certificate includes the provider public key, the provider-id (which is a hash of the provider public key), and the provider database ID.
  • the client certificate comprises the client/user public key, which then acts as the client/user ID, the client ID, and the provider computer database ID.
  • the provider servlet 56 accepts various messages from providers as will be discussed more fully below, and routes them to the appropriate provider manager 58. Further, the provider servlet 56 returns responses from the provider manager to the various provider over the network 28.
  • the web server of the present invention encrypts and decrypts requests, authenticates requests, routes them to the appropriate servlet, encrypts and returns responses.
  • the provider servlet accepts GET and POST messages from providers. (These messages will be more fully explained below).
  • Provider manager 58 manages the interaction of the system of the present invention with the various providers. The provider manager receives and routes requests from other internal server components such as the client manager 68, the task scheduler 64, and the database server 54. All administrative information and processing that relates to providers is managed by the provider manager 58.
  • the element servlet 60 is used as the point where data elements and code are exchanged between clients and providers over the network. Elements are executables, or portions of executables that are to be run on provider processors.
  • Elements are uploaded to the element servlet that then stores them on the file server.
  • the CE downloads the required elements for that task from the element servlet.
  • File server 62 stores data and executable elements that are to be exchanged between clients and providers.
  • Client servlet 66 accepts various messages from clients and routes them to the appropriate client manager 68.
  • Client servlet 66 receives responses from client manager 68 and returns the response to the client. Since responses to and from clients proceed over network 28, the client servlet is invoked by a web server that is responsible for encryption, decryption, and authentication.
  • the client servlet also manages sessions with a client. This is accomplished by the client servlet assigning a cookie to the particular client for the particular session.
  • Client manager 68 is responsible for managing requests to and from clients. Client manager 68 maintains a valid message channel for each client session. When the channel is established, the client servlet accepts GET and POST requests and forwards them to the appropriate CM. The client manager uses business routines within the client manager to handle POST requests. Depending on the type of requests, such processing may involve interaction with the database server 54, the task scheduler 64, and/or the provider manager 58. A GET request is handled by the client manager 68 by returning the appropriate message from the client output queue (explained below).
  • Task scheduler 64 manages the assignment of tasks to providers.
  • the task scheduler is responsible for tracking the states of the various providers, that is, whether the providers are currently processing tasks, whether the providers are available, and, without limitation, whether the provider has the appropriate processing power to process the specific tasks that may be necessary.
  • the task scheduler 64 receives messages from client manager 68 and provider manager 58 regarding any changes in the status of clients and providers. When tasks are assigned to specific providers, the client manager keeps a record of such task assignments in the database 54.
  • Various algorithms present in the task scheduler ensure that appropriate tasks are assigned to the appropriate providers given priority of tasking and other prioritizing data.
  • Database server 54 stores information on clients and providers.
  • the database server 54 comprises an account database relating to providers and revenue accumulated by providers, a client database relating to the identity and contact information for clients, job and task information, and a provider database relating to providers and identification and contact information for providers.
  • Such information while shown as being located on a single database server 54 can also be located on multiple servers to so that the system can be scaled up to handle many more users as well as ensure privacy and security of the information.
  • the web manager 70 intermediates access between the account section of the web site and the database. The web manager 70 supports the web based registration, update of registration and account information, and retrieval of account-specific and system- wide statistics.
  • Monitoring and control module 72 comprises a set of utilities that support the overall management of the system of the present invention.
  • the monitoring and control functionality 72 provides graphical user interfaces for system operators to monitor the health of the overall system.
  • the monitoring and control functionality 72 polls all servers for health information in the form of name- value pairs that contains status information and related attributes to all of the servers. In this fashion, the monitoring and control 72 can keep track of the health of the entire system.
  • the monitoring and control module can also respond to asynchronous notification of system events from servers.
  • Log server 74 contributes to the health of the overall system by collecting debug and error log data. Any errors that are encountered throughout the system whether they are in clients, providers, or any of the servers of the system are collected in the log server for subsequent analysis and debugging where necessary.
  • this architecture is not meant as a limitation.
  • database server 54 comprises information on clients and providers and is partitioned into two databases, one for a client database, and one for a provider database. These databases can run on a single server or on dedicated servers. As the number of clients and providers increases, additional databases can be added.
  • Client servlet 66 and provider servlet 56 are stateless, therefore allowing additional components to be easily added to increase capacity. When a client request comes in over network 28, the requests are allocated among available client servlets to increase the response. The same holds true for the provider servlet 60, thus allowing responses to and from providers to be rapidly responded to.
  • Element servlet 60 is also stateless, thereby allowing additional components to be easily added to increase capacity as well.
  • a large number of element servlets may potentially exist which store elements on a given file server.
  • Clients and providers allocate their requests and elements across the element servlets.
  • File server 62 will store large amounts of client data. Additional servers can be added to the system to increase this capacity.
  • Provider manager 58 and client manager 68 can also be augmented by additional client and provider managers respectively.
  • Each client is assigned to a specific client manager. This is accomplished by assigning clients to groups, with the group subsequently being assigned to a client manager. For example, client manager 1 may handle groups 0-49, and client manager 2 may handle groups 50-99.
  • Task scheduler 64 can quickly become a bottleneck when multiple clients and providers begin to grow. If this is the case, additional task schedulers can be added to manage a subset of tasks flowing to and from a provider. Again, tasks and providers can be assigned to groups with groups being subsequently assigned to specific task schedulers. Again, as task schedulers are added, a real allocation of tasks and providers among the various schedulers is accomplished in order to even the workflow and ensure the most efficient functioning of the system of the present invention. In a similar fashion, web manager 70 can be augmented by additional instances of web managers with requests being allocated among the various web managers.
  • the monitoring and control module 72 can also be represented by multiple instances of the monitoring and control function. There can also be multiple instances of the log server 74. However, there must be close coordination among the various log servers so that any trends and errors can be detected and corrected.
  • Registration servlet 50 can also be augmented by additional components. Clients and providers are assigned across the various instances of registration servlets.
  • the task scheduler 64 maintains an in-memory model of the state of all of the providers (i.e. the provider profiles) and any tasks that it manages. Information to establish the task scheduler is initially retrieved from database 54. It is thereafter updated as messages are received from the provider manager regarding status of efforts of the various providers, and from the client manger, which forwards relevant messages from clients to the task scheduler.
  • the task scheduler also requests additional information about providers by cuing a get task message (for retrieving task status) or a get cache contents for obtaining the cache contents of the server for the provider.
  • the task scheduler comprises various scheduling algorithms. This task scheduler makes initial assignments of tasks to providers and migrates tasks from one provider to another if necessary. When the task scheduler schedules a new task, it records the assignment of the task and the provider to which it is assigned in the task database 54. It further queues a task message to the provider. When migrating a task, the task scheduler removes the task from the current provider by cuing a remove task message, assigning the partially completed task to a new provider via a task message. All such operations are reported in the task database.
  • the element servlet is the entity that stores data elements that are to be assigned to providers for processing. Clients upload their data elements to the element servlet 60.
  • the element servlet authenticates the client as one that is permitted to store elements in the file server 62. If the client is authenticated, the element servlet 60 stores the object in a physical location on the file server 62 and notifies the client manager 68 that the object has been uploaded. The client manager 68 in the client database 54 then records this information. Client computers download elements to the element manager.
  • the element manager 60 registers the fact that this has occurred to the client manager 68.
  • the notification that is provided by the element servlet to the client manager 68 includes the client ID, a locator for the data element, and the physical location of the data element.
  • a provider When a provider needs an element, it makes a request to the element servlet 60 over the Internet 28.
  • the element servlet 60 authenticates the provider, decrypts the ticket that the provider offers to the element servlet, compares the ID of the provider making the request with the provider ID and the ticket. If the ID's match and the ticket has not expired, the element servlet 60 looks up the element on the file server 62 and returns it to the provider for execution.
  • the client terminates a particular job, all of the elements associated with that job are marked as deleted. Deleted elements are removed from the file server 62 and from the database 54. Communication within the system of the present invention among the CE, the CTS, and the client is governed by a communication protocol unique to the present invention.
  • the message channel is implemented above a low-level protocol that is part of this invention.
  • the low-level protocol supports the secure and reliable transmission of messages between system components (CE, client, CTS).
  • the protocol is secure because it is built on Secure HTTP (HTTP over SSL).
  • the invention uses SSL to both provide confidentiality (encryption) of transmitted data and to authenticate the parties at each end of the connection.
  • Providers and client applications can only access the server if they have a legitimate certificate.
  • the server uses a hash of the client's public key as a client identifier.
  • the protocol is reliable because it incorporates sequence numbers, acknowledgement, and retransmission. Combined, these protect against dropped messages, out-of-order processing of messages, and the processing of duplicate messages. It is worth noting, the low level protocol provides the abstraction of a bi-directional communication channel between components even though it is implemented on top of HTTP that has a request-response model.
  • a bi-directional model is simulated by having the client periodically "poll" the server for messages. This poll is done on an exponential decay to limit the amount of traffic generated by polling. If the CTS does not respond to a particular GET message, perhaps because of heavy CTS load, subsequent GET messages are sent ever more infrequently, according to an exponential decay, until some limit periodicity is achieved.
  • the maximum polling rate is present each time the client and server exchange a high level message.
  • the two basic operations in the low level protocol are the "GET” and the "POST", both of which are initiated by the client.
  • the POST operation sends a new message (or set of messages) to the server.
  • the GET is a poll operation to get any message (or set of messages) that the server may have enqueued for the client. Since the low-level protocol is based on HTTP it takes advantage of that protocol's features for multipart messages and different message encodings (e.g., for compression). Further, multipart messages can be sent in order to maximize communication efficiency.
  • All messages are also subject to data compression to limit the volume of data being transmitted and to conserve bandwidth.
  • the low-level protocol also supports the concept of session. Sessions are implemented using cookies and provide for the resetting of a communication channel in the event that synchronization is lost.
  • the following is a listing of high level message types used by the present invention in its communications protocol. This list is illustrative in nature. Other message types can be added to the present invention as the need arises.
  • a series of messages are transmitted between the system and various elements. These messages are characterized as Client-to-Server messages, Server-to-Provider messages, Provider-to-Server messages and Server-to-Client messages.
  • Client-to-Server Messages GetConfig This message is sent when a particular client wants to reestablish its configuration.
  • a GetConfig message causes the CTS to send a Config message to the client.
  • Task The client uses this message to send the server a task to be executed.
  • ClientProfile The client sends this message on startup and whenever client settings change.
  • Create Job Create a new job. Stop: Stop execution of a job or task. Resume: Resume execution of a stopped job or task.
  • RemoveJob Remove (cancel) a j ob .
  • GetJobProperties Requests a JobProperties message.
  • GetContents Requests a Contents message.
  • GetExecutableElement Request that an ExecutableElement message be sent describing a particular executable element.
  • GetDataElement Request that a DataElement message be sent describing a particular data element
  • RemoveExecutableElement Request the removal of the named executable element.
  • RemoveDataElement Request the removal of the named data element.
  • Attach Attach to a particular task or job.
  • the attach message is used to subscribe to receive status updates on the identified task or job.
  • GetTask Request that the identified task be checkpointed. The checkpoint is returned to the client in a Task message.
  • GetTaskStatus Requests the status of a particular task. The server will send a TaskStatus message in response.
  • RemoveTask Requests the removal of the named task.
  • ExecutableElement Register an executable element with the server. The actual executable element data must be downloaded over the data channel.
  • DataElement Register a data element with the server. The actual data element data must be downloaded over the data channel. CloseSession: Close current session. Error: Report an error condition.
  • ExternalDelivery Alert the recipient that a message is waiting at a specified URL.
  • Heartbeat Alert the recipient that the sending entity is still active.
  • Task Assigns a task to a provider.
  • GetTaskStatus Requests a task status for a particular task.
  • Provider responds with a TaskS tatus message.
  • GetTask Requests a checkpoint for a particular task.
  • the provider responds with a Task message containing the checkpoint.
  • Config Sent to change the providers server-controlled configuration settings.
  • GetContents Request contents of providers message queue.
  • WorkSummary Send current node statistics to CE for display to the user.
  • GetCacheContents Request contents of provider's cache.
  • Provider responds with a CacheContents message.
  • Display Content Send new display content to CE.
  • Upgrade Send a software upgrade to CE.
  • Error Report an error condition.
  • ExternalDelivery Alert the recipient that a message is waiting at a specified URL.
  • Reports profile information for node e.g., machine type, available disk, available memory, connection speed, operating system type.
  • Intermediate task statuses may contain partial results.
  • Final task statuses contain the results of the task.
  • Task statuses are also used to report errors in the execution of a task.
  • Task The provider sends the task message out in response to a GetTask message from the provider.
  • the Task message is a checkpoint of the state of the identified task.
  • Heartbeat Alert the recipient that the sending entity is still active.
  • Task The server sends the task message out in response to a GetTask message from the client.
  • the Task message is a checkpoint of the state of the identified task.
  • Task Status Reports the status of a task. This could be generated in response to a task status from the provider, in response to a GetStatus from the client, or due to an internal server condition.
  • Config Sent to change the clients server-controlled configuration settings.
  • JobProperties Report current job properties.
  • Contents Return contents of job or client global area.
  • ExecutableElement Return information on a specific executable element. Sent in response to a GetExecutableElement message.
  • DataElement Return information on a specific data element.
  • the present invention can also be employed in other manners, such as a method of marketing computers by offering incentives to computer customers that agree to operate a compute engine (CE) on the computers and having the CE utilize idle computational processing power on the computers. Incentives can include, but are not limited to free computer use, free ISP service, discounted computer sales price, discount computer lease price, a sales rebate, periodic rebates, and usage fees.
  • the CE can also be utilized to deliver "pushed" content, such as advertising, to these computer customers via a display window of said computer's graphic interface or via said computer's sound output.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Multi Processors (AREA)
  • Power Sources (AREA)

Abstract

L'invention concerne une plate-forme informatique distribuée utilisant la puissance de traitement informatique en veille de plusieurs « ordinateurs fournisseurs ». Au moins, un serveur réseauté recueille les tâches de plusieurs ordinateurs de clients, programme et distribue les tâches aux « ordinateurs fournisseurs » réseautés, puis recueille et renvoie les résultats aux ordinateurs des clients. Une interface de programme d'application client compose les tâches et recueille les résultats. Un moteur de calcul fonctionne sur les « ordinateurs fournisseurs », afin de communiquer avec le serveur et exécuter les tâches, en utilisant la puissance de traitement en veille.
EP00959899A 1999-08-26 2000-08-28 Procede et systeme d'etablissement et d'utilisation de la puissance de traitement informatique reseautee en veille Withdrawn EP1145109A3 (fr)

Applications Claiming Priority (5)

Application Number Priority Date Filing Date Title
US15076699P 1999-08-26 1999-08-26
US150766P 1999-08-26
US21033400P 2000-06-12 2000-06-12
US210334P 2000-06-12
PCT/US2000/024336 WO2001014961A2 (fr) 1999-08-26 2000-08-28 Procede et systeme d'etablissement et d'utilisation de la puissance de traitement informatique reseautee en veille

Publications (2)

Publication Number Publication Date
EP1145109A2 true EP1145109A2 (fr) 2001-10-17
EP1145109A3 EP1145109A3 (fr) 2002-09-11

Family

ID=26848010

Family Applications (1)

Application Number Title Priority Date Filing Date
EP00959899A Withdrawn EP1145109A3 (fr) 1999-08-26 2000-08-28 Procede et systeme d'etablissement et d'utilisation de la puissance de traitement informatique reseautee en veille

Country Status (6)

Country Link
EP (1) EP1145109A3 (fr)
JP (1) JP2003507812A (fr)
AU (1) AU7114200A (fr)
CA (1) CA2352807A1 (fr)
IL (1) IL142806A0 (fr)
WO (1) WO2001014961A2 (fr)

Families Citing this family (40)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6963897B1 (en) 2000-03-30 2005-11-08 United Devices, Inc. Customer services and advertising based upon device attributes and associated distributed processing system
US6891802B1 (en) 2000-03-30 2005-05-10 United Devices, Inc. Network site testing method and associated system
US7082474B1 (en) 2000-03-30 2006-07-25 United Devices, Inc. Data sharing and file distribution method and associated distributed processing system
US7254607B2 (en) 2000-03-30 2007-08-07 United Devices, Inc. Dynamic coordination and control of network connected devices for large-scale network site testing and associated architectures
US6654783B1 (en) 2000-03-30 2003-11-25 Ethergent Corporation Network site content indexing method and associated system
USRE42153E1 (en) 2000-03-30 2011-02-15 Hubbard Edward A Dynamic coordination and control of network connected devices for large-scale network site testing and associated architectures
US7020678B1 (en) 2000-03-30 2006-03-28 United Devices, Inc. Machine generated sweepstakes entry model and associated distributed processing system
US20010039497A1 (en) * 2000-03-30 2001-11-08 Hubbard Edward A. System and method for monitizing network connected user bases utilizing distributed processing systems
US7039670B2 (en) 2000-03-30 2006-05-02 United Devices, Inc. Massively distributed processing system with modular client agent and associated method
US20090216641A1 (en) 2000-03-30 2009-08-27 Niration Network Group, L.L.C. Methods and Systems for Indexing Content
US8010703B2 (en) 2000-03-30 2011-08-30 Prashtama Wireless Llc Data conversion services and associated distributed processing system
US7003547B1 (en) 2000-03-30 2006-02-21 United Devices, Inc. Distributed parallel processing system having capability-based incentives and associated method
AU2001263056A1 (en) * 2000-05-12 2001-11-26 Datasynapse, Inc. Methods, apparatus, and articles-of-manufacture for network-based distributed computing
US7102796B2 (en) * 2001-08-23 2006-09-05 Hewlett-Packard Development Company, L.P. Network printing system using a process power broker for directing print job from busy machines to idle machines for processing only before printing
EP1329809B1 (fr) * 2002-01-18 2008-08-06 Hewlett-Packard Company, A Delaware Corporation Système et méthode informatique distribuée
US7143411B2 (en) * 2002-03-15 2006-11-28 Hewlett-Packard Development Company, L.P. Capping processor utilization
US8020162B2 (en) * 2002-07-19 2011-09-13 International Business Machines Corporation Employing a resource broker in managing workloads of a peer-to-peer computing environment
FR2857472B1 (fr) * 2003-07-10 2007-01-26 Univ Reims Champagne Ardenne Systeme de gestion distribuee des ressources informatiques et des calculs
US7197433B2 (en) * 2004-04-09 2007-03-27 Hewlett-Packard Development Company, L.P. Workload placement among data centers based on thermal efficiency
US7765552B2 (en) 2004-09-17 2010-07-27 Hewlett-Packard Development Company, L.P. System and method for allocating computing resources for a grid virtual system
US20060080389A1 (en) * 2004-10-06 2006-04-13 Digipede Technologies, Llc Distributed processing system
GB2419693A (en) * 2004-10-29 2006-05-03 Hewlett Packard Development Co Method of scheduling grid applications with task replication
US20070033247A1 (en) * 2005-08-02 2007-02-08 The Mathworks, Inc. Methods and system for distributing data to technical computing workers
US7917573B2 (en) * 2005-11-30 2011-03-29 International Business Machines Corporation Measuring and reporting processor capacity and processor usage in a computer system with processors of different speed and/or architecture
CN101331460B (zh) * 2005-12-16 2012-10-03 国际商业机器公司 基于处理能力的实时基准对软件产品的使用进行计量的方法和系统
US20070226226A1 (en) * 2006-03-23 2007-09-27 Elta Systems Ltd. Method and system for distributing processing of computerized tasks
ITUD20060108A1 (it) * 2006-04-21 2007-10-22 Xteque S R L Apparato e procedimento per condividere e/o rendere accessibile la potenza di calcolo computazionale fra calcolatori connessi in rete
US7724149B2 (en) * 2007-06-11 2010-05-25 Hewlett-Packard Development Company, L.P. Apparatus, and associated method, for selecting distribution of processing tasks at a multi-processor data center
GB2468169A (en) * 2009-02-28 2010-09-01 Geoffrey Mark Timothy Cross A grid application implemented using a virtual machine.
US20120124193A1 (en) 2010-11-12 2012-05-17 International Business Machines Corporation Identification of Critical Web Services and their Dynamic Optimal Relocation
US9658901B2 (en) * 2010-11-12 2017-05-23 Oracle International Corporation Event-based orchestration in distributed order orchestration system
CN102480512B (zh) * 2010-11-29 2015-08-12 国际商业机器公司 用于扩展服务器端处理能力的方法和装置
US20140304713A1 (en) * 2011-11-23 2014-10-09 Telefonaktiebolaget L M Ericsson (pulb) Method and apparatus for distributed processing tasks
AU2013315670A1 (en) * 2012-09-11 2015-04-09 Ajay JADHAV Platform agnostic modular framework
WO2016040889A1 (fr) * 2014-09-11 2016-03-17 Cisco Technology, Inc. Cloud computing à externalisation ouverte
US10873540B2 (en) 2016-07-06 2020-12-22 Cisco Technology, Inc. Crowd-sourced cloud computing resource validation
US10360606B2 (en) 2016-07-19 2019-07-23 Cisco Technology, Inc. Crowd-sourced cloud computing in a multiple resource provider environment
CN111106946B (zh) * 2018-10-29 2022-12-30 金山云(深圳)边缘计算科技有限公司 一种联网设备边缘计算节点改造方法和系统
CN111061553B (zh) * 2019-12-17 2023-10-10 湖南大学 一种用于超级计算中心的并行任务调度方法和系统
US20230117047A1 (en) * 2021-10-15 2023-04-20 Microsoft Technology Licensing, Llc Systems and methods for allocating compute nodes in a power-constrained environment

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See references of WO0114961A3 *

Also Published As

Publication number Publication date
WO2001014961A3 (fr) 2001-08-02
IL142806A0 (en) 2002-03-10
CA2352807A1 (fr) 2001-03-01
EP1145109A3 (fr) 2002-09-11
AU7114200A (en) 2001-03-19
WO2001014961A2 (fr) 2001-03-01
JP2003507812A (ja) 2003-02-25

Similar Documents

Publication Publication Date Title
US6463457B1 (en) System and method for the establishment and the utilization of networked idle computational processing power
EP1145109A2 (fr) Procede et systeme d'etablissement et d'utilisation de la puissance de traitement informatique reseautee en veille
US20210255986A1 (en) Managing use of program execution capacity
US9559938B2 (en) Method, system and apparatus for providing pay-per-use distributed computing resources
CN106055412B (zh) 用于动态管理计算容量请求的方法和系统
US7533170B2 (en) Coordinating the monitoring, management, and prediction of unintended changes within a grid environment
Lai et al. Tycoon: An implementation of a distributed, market-based resource allocation system
US20110041126A1 (en) Managing workloads in a virtual computing environment
JP4965460B2 (ja) グリッド・ジョブに関するグリッド・プロバイダの選択を自動的に制御するための方法
US20180095997A1 (en) Secure automated resource-exchange system
US8346591B2 (en) Automating responses by grid providers to bid requests indicating criteria for a grid job
US6857020B1 (en) Apparatus, system, and method for managing quality-of-service-assured e-business service systems
US10616139B1 (en) Reducing quota access
US20120221454A1 (en) Systems and methods for generating marketplace brokerage exchange of excess subscribed resources using dynamic subscription periods
US20060168584A1 (en) Client controlled monitoring of a current status of a grid job passed to an external grid environment
JP2004513406A (ja) 大量分散処理システムおよび関連する方法
US11748168B2 (en) Flexible batch job scheduling in virtualization environments
Högberg An applied evaluation and assessment of cloud computing platforms

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

AK Designated contracting states

Kind code of ref document: A2

Designated state(s): AT BE CH CY DE DK ES FI FR GB GR IE IT LI LU MC NL PT

AX Request for extension of the european patent

Free format text: AL;LT;LV;MK;RO;SI

XX Miscellaneous (additional remarks)

Free format text: DERZEIT SIND DIE WIPO-PUBLIKATIONSDATEN A3 NICHT VERFUEGBAR.

PUAK Availability of information related to the publication of the international search report

Free format text: ORIGINAL CODE: 0009015

AK Designated contracting states

Kind code of ref document: A3

Designated state(s): AT BE CH CY DE DK ES FI FR GB GR IE IT LI LU MC NL PT

AX Request for extension of the european patent

Free format text: AL;LT;LV;MK;RO;SI

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE APPLICATION IS DEEMED TO BE WITHDRAWN

18D Application deemed to be withdrawn

Effective date: 20020205