WO2008133818A1 - Distributed, fault-tolerant and highly available computing system - Google Patents
Distributed, fault-tolerant and highly available computing system Download PDFInfo
- Publication number
- WO2008133818A1 WO2008133818A1 PCT/US2008/004866 US2008004866W WO2008133818A1 WO 2008133818 A1 WO2008133818 A1 WO 2008133818A1 US 2008004866 W US2008004866 W US 2008004866W WO 2008133818 A1 WO2008133818 A1 WO 2008133818A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- component
- message
- components
- deterministic
- execution
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
-
- 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
- G06F9/54—Interprogram communication
- G06F9/546—Message passing systems or structures, e.g. queues
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
- G06F11/14—Error detection or correction of the data by redundancy in operation
- G06F11/1402—Saving, restoring, recovering or retrying
- G06F11/1446—Point-in-time backing up or restoration of persistent data
- G06F11/1458—Management of the backup or restore process
- G06F11/1461—Backup scheduling policy
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
- G06F11/14—Error detection or correction of the data by redundancy in operation
- G06F11/1479—Generic software techniques for error detection or fault masking
- G06F11/1482—Generic software techniques for error detection or fault masking by means of middleware or OS functionality
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F15/00—Digital computers in general; Data processing equipment in general
- G06F15/16—Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
- G06F11/14—Error detection or correction of the data by redundancy in operation
- G06F11/1402—Saving, restoring, recovering or retrying
- G06F11/1446—Point-in-time backing up or restoration of persistent data
- G06F11/1458—Management of the backup or restore process
- G06F11/1469—Backup restoration techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
- G06F11/14—Error detection or correction of the data by redundancy in operation
- G06F11/1479—Generic software techniques for error detection or fault masking
- G06F11/1482—Generic software techniques for error detection or fault masking by means of middleware or OS functionality
- G06F11/1484—Generic software techniques for error detection or fault masking by means of middleware or OS functionality involving virtual machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2201/00—Indexing scheme relating to error detection, to error correction, and to monitoring
- G06F2201/835—Timestamp
Definitions
- the present invention relates generally to application development tools, methods and systems for developing and executing distributed systems, and more particularly, an improved environment for developing and executing distributed, transparently fault-tolerant, and highly available systems for executing component oriented applications.
- SCA Service Component Architecture provides an open, technology-neutral model for implementing IT services that define a business function.
- the SCA also provides a model for the assembly of business solutions from collections of individual services, with control over aspects of the solution such as access methods and security.
- customers are able to more easily create new, and transform existing, IT assets into reusable services that may be rapidly adapted to changing business requirements.
- the specifications that enable the building of SCA (middleware) components take advantage of a Service-Oriented Architecture (SOA), which structures IT assets as a series of reusable services that perform business functions.
- SOA Service-Oriented Architecture
- the Service Oriented Architecture combines the ability to invoke remote objects and functions (called "services") with tools for dynamic service discovery, placing an emphasis on interoperability.
- services remote objects and functions
- a current solution implements entity beans, e.g., "Enterprise Java Bean” (EJB) that include the server-side component architecture for the J2EE platform.
- EJBs purportedly support rapid and simplified development of distributed, transactional, secure and portable Java applications.
- EJBs support a container architecture that allows concurrent consumption of messages and provide support for distributed transactions, so that database updates, message processing, and connections to enterprise systems using the J2EE architecture can participate in the same transaction context.
- the deterministic computing system comprises: a plurality of software components each implementing logic to perform a task, the components executing in an execution environment comprising on one or more machines connected in a network and adapted for communicating messages between the components; means for automatically associating a deterministic timestamp in conjunction with a message to be communicated from a sender component to a receiver component during program execution, said timestamp representative of estimated time of arrival of said message at a receiver component; means for deterministically executing said component by using said timestamps to generate a unique arrival order of input messages; and, means for tracking state of a component during program execution, and periodically checkpointing the state to a local storage device; wherein upon failure of a machine, the component state is restored by recovering a recent stored checkpoint and re-executing the events occurring since the last checkpoint.
- the computing system is deterministic by repeating the execution of a receiving component by processing the messages in the same order as their associated timestamps.
- the component state may be checkpointed to a backup processor device.
- a method for deterministic execution of components in a computing system adapted for enabling message communication amongst and between said components, each component implementing logic to perform a task, the method comprising: automatically associating a timestamp in conjunction with a message to be communicated from a sender component to a receiver component during program execution, the timestamp representative of estimated time of arrival of the message at a receiver component; deterministically executing said component by using said timestamps to generate a unique arrival order of input messages; and, tracking state of that component during program execution, and periodically checkpointing the state to a local storage device; wherein upon failure of a component, the component state is restored by recovering a recent stored checkpoint and re-executing the events occurring since the last checkpoint.
- an environment for executing component-oriented applications the environment adapted for communicating messages amongst and between components.
- the environment comprises: a sub-system enabling high-level design of the components including specification of ports representing message inputs to and outputs from the components, each component implementing logic to perform a task; a placement service for mapping a high-level design onto a lower level implementation in which components are assigned to particular processing engines of said execution environment; a code enhancing means for: automatically augmenting an input or output message of an associated component with a timestamp in conjunction with a message to be communicated from a sender component to a receiver component during program execution, the timestamp representative of estimated time of arrival of the message at a receiver component; and, augmenting a component with code for tracking incremental changes to its state; wherein, a state of a component is tracked during program execution, and the component state checkpointed in a local storage device; and, wherein upon failure of a component, the component state is restored by recovering
- the high-level design is neutral with respect to where components execute in said execution environment.
- Figure 1 depicts an execution server architecture 10 in which a middleware application is run for executing components and component oriented applications in accordance with the present invention
- Figure 2 depicts example messaging and communication between components in a single execution engine - in this example a Java Virtual Machine(JVM), and, between and among multiple execution engines according to the execution server architecture of the present invention
- Figures 3A-3D depict an example application enabled by the server middleware components of the present invention, including the placement of components from an example user-design (Figs. 3 A, 3B), virtual time estimation (Fig. 3C), and, silence and curiosity message generation (Fig. 3D);
- Figure 4 demonstrates an overview of checkpointing to a memory storage device and checkpoint recovering from a hard disk according to the principles of the invention
- Figures 5A-5C demonstrates an overview of using passive remote backups and spawning scheduler replicas for providing the high availability afforded by the present invention.
- the proposed invention aims to address the problems in the art, namely the continued need to provide programmers with specialized methodologies and structures such as transactions, JDBC, or entity beans that separate out component state into separate objects in order to persist that state, and the need to log messages between components in non-deterministic implementations.
- This is addressed by providing a deterministic and highly available execution server that automatically provides persistence and fault-tolerance for executing component oriented applications.
- FIG. 1 depicts an execution server architecture 10 in which the present middleware application is run for executing components and component oriented applications, represented generally as components 20a, . . ., 2On.
- each server includes one or more Java Virtual Machines, represented generally as JVM' s 12a,. . . 12n, such as shown in Fig. 1.
- JVM' s 12a . . . 12n
- JVMs Java Virtual Machine
- FIG. 1 depicts an execution server architecture 10 in which the present middleware application is run for executing components and component oriented applications, represented generally as components 20a, . . ., 2On.
- JVM' s 12a represented generally as JVM' s 12a,. . . 12n, such as shown in Fig. 1.
- JVMs Java Virtual Machine
- a scheduler layer containing at least one scheduler, represented generally as schedulers 15a,. . . 15n, as a layer between JVMs and components in the hierarchy that are transparent to developers.
- a scheduler 15a, 15b . . . 15n is a grouping of one or more tightly related components in a JVM, together with logic deciding when to execute these components.
- a second JVM 13 includes scheduler 17 for managing components 28 and 29.
- the scheduler layer between JVMs and components in the hierarchy manages the execution of components in a JVM. It is understood that, within a JVM, the communication between components and that between schedulers are efficiently implemented by reference passing. There may be multiple schedulers in the JVM, each managing the execution of its own components. When a component has an input message pending, the scheduler will choose an appropriate time to assign an available thread to execute it. This decision is made based on a number of factors, such as system load, user-required throughput, memory consumption caused by queuing, etc. Concurrency is obtained when the scheduler decides to execute multiple components at the same time.
- a component may be service-oriented or event-oriented and may be any collection of "objects" that are consistent with an SCA-like component model.
- the Java Language or C++ Language or like object-oriented language, or other languages such as Python or Perl are used for implementing SCA service components, and the data sent between components. That is, interaction between components occurs only by passing data messages across ports, or by a service call from a service consumer to a service provider, in which data values can be passed and returned.
- Receivers of sent messages or service calls appear as objects with synchronized methods.
- One thread of control exists within a component at any one time. No object is ever shared between components.
- Objects are either: (a) the component object itself, a "monitor” with synchronized methods, (b) "value” objects, that can be passed from component to component, but never shared, or (c) "implementation” objects, that can be shared, but only within either the component object itself or within the same value object.
- This discipline which assures, among other things, that no data is concurrently owned by more than one executing component, is formalized and described in a reference authored by David Bacon, Robert Strom, Ashis Tarafdar entitled “Guava: a dialect of Java without data races," Proceedings of the 15th ACM SIGPLAN conference on Object-oriented programming, systems, languages, and applications, 2000 the contents and disclosure of which is incorporated by reference as if fully set forth herein. According to the Guava techniques, it is possible to statically check that a particular component obeys this discipline.
- the Guava dialect of Java, running on a JVM is just one example of the kind of component implementation language suitable for use in the environment of this invention.
- the execution server of the invention is designed to support distributed execution across machines. Multiple instances of Java Virtual Machine (JVMs) may run on those machines or run as different processes within a single machine, depending on the deployment.
- JVMs Java Virtual Machine
- UDP User Datagram Protocol
- the JVM 12 communicates via UDP messaging protocol with the JVM 13.
- TCP Transmission Control Protocol
- loss-free and order-preserving messaging is built in the middleware component of the invention as will be described in greater detail herein below.
- the Execution Server of the invention is distinguished from other approaches because: (1) the development environment is radically simplified: Components can be written in plain Java, and can be wired, as in SCA component models, to build a distributed flow network; (2) Components store state in plain Java primitive variables and collection classes; (3) Programmers need not use specialized methodologies and structures, such as transactions, JDBC, or entity beans to separate out component state into separate objects and to persist that state. Instead, persistence and fault-tolerance are provided automatically by the execution server(s) of the invention, despite machine failures, and faulty networks that may drop, reorder, or duplicate messages.
- the technology for fault- tolerance is based upon assuring deterministic, replayable execution of the system. This is an improvement over other approaches based upon transactions, or based on replication, or based on persistently logging all inter-component messages.
- Deterministic execution means that if the system is given the same state and the same input messages, it will generate the same output messages. Determinism implies that upon a failure of a component, its state can be restored by recovering a recent checkpoint, and replaying the events occurring since that checkpoint. Because of determinism, the state after the replay is guaranteed to be the same as the lost state. This means that state does not need to be saved persistently each time it is updated, but only intermittently.
- each message communicated between components is tagged with a virtual time (VT).
- VT virtual time
- the virtual time is chosen to be a deterministic estimate of the real time at which the message will arrive at the receiving component.
- the execution server of the invention guarantees that the system will behave as if the messages had arrived in virtual time order. The better the correspondence between virtual time and real time, the better the system will perform; however, deterministic execution is guaranteed regardless of how good this correspondence is.
- the virtual time is a real-time deadline, and the system will adjust the scheduling to guarantee to meet these real-time deadlines. This is also unlike the case of event simulators, wherein simulated virtual time may have no obvious correlation with real time.
- Deterministic execution is achieved by augmenting all communications with timestamps containing virtual times (VTs).
- VTs virtual times
- the system without manual intervention, automatically assigns VTs to those events.
- the VTs conform to causal order, i.e., if an output is caused by an input, it must not occur at an earlier virtual time, and is an approximation to the real time.
- a log records all the assigned VTs, because their generation is non-deterministic, and the log will be essential when replay is required later.
- a component When a component is chosen to execute, it may produce outputs as a result of processing the input message.
- the outputs are associated with VTs strictly greater than the input message's VT, reflecting the non-zero computation delay.
- the increment in VT, as well as the output messages, is fully determined by the input message.
- a component may receive inputs, for example, sent messages, or service call messages, etc. from multiple predecessor components. In this case, a deterministic merge is utilized based on VTs of the messages. VTs are generated independently by the predecessors, but they are compared locally at the merge. The message with the smallest VT can be safely processed once it is known that no predecessor can send any message in the future with an earlier VT.
- messages can be processed aggressively and rolled back if a message with earlier VT arrives, as in the discrete event simulation environment described in the reference to Jefferson, D. entitled “Virtual time”, ACM Transactions on Programming Languages and Systems, July 1985. Because of deterministic VT generation and deterministic merging (both of which are local to the components), the order of message processing is deterministic.
- the components' computation logic may be written in plain Java, C++ or like objected-oriented code, or in other languages such as Python or Perl. Ports are defined, for example, with Java interfaces. Logical connections are created by the designers by wiring two or more ports.
- This high-level design completely abstracts away the distributed nature of the resulting system. It also takes such properties as deterministic execution and fault tolerance for granted.
- a middleware placement service component 50 is used to map this high-level design such as the design shown in Fig. 3 A, onto lower-level implementation, where the most inter-related components are grouped into schedulers, and schedulers are assigned in JVMs, possibly on different machines.
- the placement before execution is semi-automatic, with user-supplied parameters such as metrics of machines' computation power, network capability, and placement preferences. To maximize performance, work load is balanced on different machines.
- the initial placement may be fine-tuned by moving schedulers from their original machines to other machines for even better overall performance.
- logical connections in the developers' view may also be mapped to physical connections.
- This placement service 50 simplifies the deployment task, but still provides enough flexibility by accepting user-specified configurations.
- Fig. 3 A there is depicted a high-level approach of an example designer's view of a ticketing (ticket selling) system 30 that a may used to model the receipt of messages from two different requesting paths, i.e., requestors 32, 34 for reserving seats at an event, for example.
- tickets requests are first received and processed to determine an amount of tickets being requested by each requestor (e.g., 5 tickets or 10 tickets) and then, finally assigned a process for printing the tickets at a printer device.
- the following example code portion depicts the components' logic (e.g., primitive variables, collection classes, etc.) that may be written in plain Java according to the high-level design of the example system described in Fig. 3 A.
- the placement service 50 will generate code for the low-level implementation as shown in Fig. 3B.
- the requestors are modeled by defining ports for a first JVM 62 which receives the messages; and, further modeled are the respective ports defined at a second JVM 64 which processes the respective requests and merges the requests at an assigner component 65 that assigns the printer function component which is modeled by defining a single port for a third JVM 66 which includes a printer component 67 for performing the scheduled ticket printing process.
- the following example code portion depict the placement logic that may be written in plain Java according to the lower-level design of the example system shown in Fig. 3B.
- This following example code depicts the configuration of the first JVM 62, which as shown in Fig. 3B is designed to include a first scheduler 72 for managing requestor components 32, 34.
- This following example code depicts the configuration of the second JVM 64, which as shown in Fig. 3 B is designed to include a scheduler 74 for managing processing components corresponding requestor components 32, 34.
- This following example code depicts the configuration of the third JVM 66, which as shown in Fig. 3 B is designed to include a scheduler 76 for managing printing processing component from a single input.
- a program transformer augments the user- written component to produce a component compatible with the run-time execution environment.
- the program transformer supplies the following enhancements: 1) All interfaces of input and output messages or method calls are augmented with a field that holds the VT; 2) Each method that processes a message arriving at an input port is augmented with an estimator that computes the "delta- VT" for each output message it generates, and for the return from that method.
- the delta- VT represents a deterministic estimate of the amount of real-time that would elapse from the start of the method to either the generated output message or to the return; and, 3)
- Each component is augmented with code that tracks incremental changes to its state since the last soft-checkpoint, and which upon request from the scheduler, serializes an incremental soft-checkpoint record.
- Soft checkpoints are so called, because any single checkpoint can be lost without compromising the system's ability to recover - a lost checkpoint merely means that recovery must proceed from the previous checkpoint, which may lengthen the time to recover after a failure, but will not affect the eventual ability to recover.
- the component is augmented with code that reconstructs a state from a collection of incremental soft- checkpoint records.
- a component may be augmented with code that generates "eager silences". That is, given that it is now known that no input messages are arriving on its input ports through a given time t, it computes the earliest delta- VT beyond time t for which it is possible for a message to appear on given output ports.
- Such a computation can be used for the purpose of sending silences to components connected to these input ports. Since a range of silent timestamps promises that no messages will ever be sent from that component with those timestamps, such information may possibly enable receiving components to proceed to process a waiting message because it is now known to be the earliest possible message.
- Fig. 3 C depicts in more detail the determinism of VT used for augmenting the message structures required for implementing the invention as depicted by the low-level design in the example ticketing system described in the example of Fig. 3B. It is understood that messages or events ("e") are associated with VTs on the system boundary. Currently VTs are generated in an arbitrary way, consistent with causality. However, they may be an estimation of the real time. Thus, as shown in Fig.
- Fig. 3 C additionally depicts the tracking of incremental changes to its state since the last soft-checkpoint, which upon request from the scheduler, causes generating an incremental soft-checkpoint record depicted by entry into a log 99 in a data storage device. This log 99 records all the assigned VTs, because their generation is non-deterministic, and the log will be essential when replay is required as will be described in greater detail herein below.
- output from the assigner component of the second JVM is a deterministically merged message sequence 55 comprising temporally sequenced messages 52 and 54, i.e., [ ] and ].
- the deterministic merger implemented in the third JVM 66 will then sequence operations in a deterministic order for the printing component 67 based on the merged VT values in received message 55. Tolerance of Message Loss and Reordering
- Intra-JVM communication is also lossless and order- preserving.
- UDP communication may be used even though this may cause message loss or reordering, because such loss or reordering is tolerated by the middleware component of the invention that implements concept of silence and curiosity messages that are sent between schedulers but not within schedulers.
- Fig. 3D there is shown the incorporation of silence and curiosity messages to ensure detection of missing messages in the inventive system.
- a time gap exists in VT, that is, unless the two messages are exactly next to each other in the time line.
- the existence of the time gap is detected by the server's middleware component, e.g., scheduler.
- silence message 81 For example, after the first JVM 62 sends the first augmented message 42, and prior to sending the second augmented message, the JVM 62 generates and sends one or more silences, i.e., silence message 81, to fill the gap so that the receiver, e.g., JVM 64, gets continuous knowledge of either messages or silences.
- a silence message defines a pair of numbers designating a range of time ticks of VT that are known not to contain messages.
- the message 80 in Figure 3D sent just before the message 42 may contain the data [Silence; 100 : -1], meaning that all ticks of VT in the range from 100 to vt, 1 -1 are known not to contain messages.
- receivers can account for every tick as either definitely containing a message or a silence. If a data or silence message is lost, the receiver will notice a gap. Gap detection is performed on the receiver side. If a gap exists for a certain period (defined as a user- configurable or programmable parameter), the receiver, e.g., JVM 64 may decide that some messages or silences have been lost, and then issue a curiosity message 83 for receipt back to the sender, e.g., JVM 62.
- a curiosity message defines a pair of numbers designating a range of time "ticks" of VT for which the receiver has no knowledge of data or silence, and is therefore "curious" about.
- the VT generation is allowed to be arbitrary provided that it is fully deterministic and it preserves the message causality. Any such VT assignment guarantees deterministic execution. However, if VTs are too far out-of-sync with real time, then a performance penalty is seen, because at a merge, the receiver would refuse to process the next message until it is sure that it will not receive any message earlier in VT. This may cause the messages generated earlier in real time to be processed later because they acquire larger VTs due to inaccuracy of the estimation. An automatic VT estimator is provided to attempt to maintain VTs and real time in approximate synchronization so that this pessimism does not cause unacceptable performance loss.
- the automatic VT estimator estimates the computation delay on individual components, based on the known complexity of the code. At deployment time, the estimator may also take into account known properties of the environment on which the component was deployed. This will adjust the computation delay estimate to reflect things such as network latency, average CPU load, garbage collector performance, expected memory consumption, threading, etc. However, it should be emphasized that the computation delay estimate must be a deterministic, repeatable function of the component state. It may not take into account non-deterministic factors, such as the actual current CPU load. If, unfortunately, a scheduler notices that the disparity between VT and real time becomes too large, it can take one of two possible actions:
- the scheduler may change its priority relative to other schedulers in the machine so that it slows down or speeds up, and reduces the disparity.
- the scheduler may make a non-deterministic decision - a so-called "determinism fault” — to adjust the parameters to the estimators used within the scheduler.
- This adjustment being non-deterministic, is allowed, unlike the normal computation delay estimate, to take into account non-deterministic factors. Because such a decision violates determinism, this action must be logged in stable storage in order to guarantee proper replay. In effect, any replay of VTs prior to the decision must use the old estimator, and any replay of VTs subsequent to the decision must use the new estimator.
- a determinism fault may result in the change to particular time estimates for particular messages.
- the goal of the system is to make the static estimators good enough so that drift between VT and real time is minimized and can be controlled by adjusting scheduling priorities, so that determinism faults are extremely rare events.
- the introduction of timestamps is automated with the VT estimator of the inventive middleware.
- some applications may need to be time-aware; or, in an even more complex scenario, certain applications may contain a mixture of non-time-aware components and time-aware components.
- the middleware component of the present invention allows timing constraints to be specified by the designers, which are automatically compiled into extra time controlling code. At run time, this code enables the schedulers to satisfy the time constraints by prioritizing the components, or moving components. That is, real-time schedulers may be programmed to work together with the placement service to guarantee real-time deadlines.
- machines may be shut down or disconnected from the network unexpectedly, or may fail.
- Many contemporary applications take advantage of the collaboration among machines.
- the use of multiple machines enables high availability. With this property, the failure of some machines does not interfere with the application's overall functionality.
- fault tolerance is transparent.
- the middleware component intermittently creates soft checkpoints for individual schedulers.
- a soft checkpoint is a compact image of the scheduler's state, including the components in it, the established connections, the queued messages, waiting service calls, etc.
- Soft checkpoints may either contain full state, or may contain incremental changes since the previous soft checkpoint.
- the state of user-defined components including, for example, language-level entities such as Java primitive values and objects with cross references
- checkpoint information 84, 86 for each of the schedulers in JVM' s 64, 66 are stored and intermittently or incrementally updated on a local memory storage device, e.g., hard disk drive 94, 96, respectively.
- a local memory storage device e.g., hard disk drive 94, 96
- checkpoints may be stored at or on remote machines.
- a JVM machine may crash, and due to this event, the schedulers running on it stop and completely lose their state since their last checkpoints. When the machine restarts, it recreates the schedulers with their last checkpoints.
- stored checkpoint information 84' for the example scheduler in JVM 64 will be communicated from the local memory storage device, e.g., hard disk drive 94, back to the restarted scheduler component executing at JVM 64.
- the schedulers will thus independently send curiosity messages 85 to their predecessors, causing the predecessors to resend the missing messages. According to the invention, the missing messages will then be reprocessed, and due to the deterministic execution afforded by the present invention, the same results will be generated.
- a scheduler catches up with the time at which it failed, its state becomes identical to its state right before it failed. From then on, it continues to process new messages.
- checkpoint information for each of the schedulers in JVM' s may be stored and intermittently or incrementally on a remote machine (e.g., JVM).
- a remotely located scheduler may perform a "passive backup" by storing checkpoints from another scheduler. If passive backups are used to store the checkpoints, then when a failure is detected, the passive backup creates instances of the backed up components, i.e., spawns a replica, and becomes active, resuming the failing scheduler's work until it restarts and catches up with the missing computation.
- Passive backups unlike active replicas of some other fault-tolerant systems, do not perform redundant computations. They merely hold checkpointed state, so that if the active machine fails, the backup is able to rapidly take over the computation with minimal delay.
- soft checkpoint information 84 for the example scheduler in JVM 64 will be communicated to a memory associated with a back-up scheduler component 74 for the second JVM 64 that is remotely stored in the third JVM 66 for storage thereat; likewise, as shown in Fig. 5A, soft checkpoint information 86 for the example third scheduler in JVM 66 will be communicated to a memory associated with a back-up scheduler component 76 for the third JVM 66 that is remotely stored in the second JVM 64 for storage thereat.
- a replica of the scheduled that was executing in JVM 64 may be generated as shown in Fig. 5B. That is, in JVM 66, a replica of the scheduled that was executing in JVM 64 may be spawned based on the remote passive back-ups. That is, a new scheduler scheduler2' is spawned that includes the components of the second JVM 64 for execution. Based on a maximum failure detection delay and the time for replica creation, the maximum downtime can be estimated in case of one failure which provides high availability assurance.
Abstract
Description
Claims
Priority Applications (5)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2010506208A JP5373770B2 (en) | 2007-04-26 | 2008-04-15 | Deterministic computing systems, methods, and program storage devices (distributed, fault tolerant, and high availability computing systems) to achieve distributed, fault tolerant, and high availability |
AU2008244623A AU2008244623B2 (en) | 2007-04-26 | 2008-04-15 | Distributed, fault-tolerant and highly available computing system |
CN2008800129532A CN101663660B (en) | 2007-04-26 | 2008-04-15 | Distributed, fault-tolerant and highly available computing system |
KR1020097016949A KR101126979B1 (en) | 2007-04-26 | 2008-04-15 | Distributed, fault-tolerant and highly available computing system |
EP08742917.1A EP2156307B1 (en) | 2007-04-26 | 2008-04-15 | Distributed, fault-tolerant and highly available computing system |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US11/740,556 US7937618B2 (en) | 2007-04-26 | 2007-04-26 | Distributed, fault-tolerant and highly available computing system |
US11/740,556 | 2007-04-26 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2008133818A1 true WO2008133818A1 (en) | 2008-11-06 |
Family
ID=39888475
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US2008/004866 WO2008133818A1 (en) | 2007-04-26 | 2008-04-15 | Distributed, fault-tolerant and highly available computing system |
Country Status (7)
Country | Link |
---|---|
US (1) | US7937618B2 (en) |
EP (1) | EP2156307B1 (en) |
JP (1) | JP5373770B2 (en) |
KR (1) | KR101126979B1 (en) |
CN (1) | CN101663660B (en) |
AU (1) | AU2008244623B2 (en) |
WO (1) | WO2008133818A1 (en) |
Families Citing this family (39)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070174695A1 (en) * | 2006-01-18 | 2007-07-26 | Srinidhi Varadarajan | Log-based rollback-recovery |
US8327327B2 (en) * | 2007-03-21 | 2012-12-04 | Carnegie Mellon University | Method for statically checking an object-oriented computer program module |
US8566829B1 (en) * | 2008-01-30 | 2013-10-22 | Emc Corporation | Cooperative multi-level scheduler for virtual engines |
US7979844B2 (en) * | 2008-10-14 | 2011-07-12 | Edss, Inc. | TICC-paradigm to build formally verified parallel software for multi-core chips |
US8060780B2 (en) * | 2008-12-16 | 2011-11-15 | Sap Ag | Transient transaction execution with on-demand persistency |
JP2010224935A (en) * | 2009-03-24 | 2010-10-07 | Olympus Corp | Distributed processing system |
US20110078666A1 (en) * | 2009-05-26 | 2011-03-31 | University Of California | System and Method for Reproducing Device Program Execution |
US8468388B2 (en) * | 2010-04-20 | 2013-06-18 | International Business Machines Corporation | Restoring programs after operating system failure |
US8782434B1 (en) | 2010-07-15 | 2014-07-15 | The Research Foundation For The State University Of New York | System and method for validating program execution at run-time |
US8724645B2 (en) | 2010-09-28 | 2014-05-13 | Microsoft Corporation | Performing computations in a distributed infrastructure |
US8516032B2 (en) * | 2010-09-28 | 2013-08-20 | Microsoft Corporation | Performing computations in a distributed infrastructure |
US20120221884A1 (en) * | 2011-02-28 | 2012-08-30 | Carter Nicholas P | Error management across hardware and software layers |
US9202185B2 (en) | 2011-04-08 | 2015-12-01 | Ca, Inc. | Transaction model with structural and behavioral description of complex transactions |
US8782614B2 (en) * | 2011-04-08 | 2014-07-15 | Ca, Inc. | Visualization of JVM and cross-JVM call stacks |
US8745434B2 (en) | 2011-05-16 | 2014-06-03 | Microsoft Corporation | Platform for continuous mobile-cloud services |
US20130124174A1 (en) * | 2011-06-03 | 2013-05-16 | David R. Jefferson | Internal parallelism in a parallel discrete event simulation for space situational awareness |
EP2608061B1 (en) * | 2011-12-22 | 2021-07-28 | Amadeus | Method and system for update message tracking and checking |
US9122595B2 (en) * | 2012-01-13 | 2015-09-01 | NetSuite Inc. | Fault tolerance for complex distributed computing operations |
US9122873B2 (en) | 2012-09-14 | 2015-09-01 | The Research Foundation For The State University Of New York | Continuous run-time validation of program execution: a practical approach |
US9069782B2 (en) | 2012-10-01 | 2015-06-30 | The Research Foundation For The State University Of New York | System and method for security and privacy aware virtual machine checkpointing |
US9563452B2 (en) | 2013-06-28 | 2017-02-07 | Sap Se | Cloud-enabled, distributed and high-availability system with virtual machine checkpointing |
US9569517B1 (en) | 2013-11-27 | 2017-02-14 | Google Inc. | Fault tolerant distributed key-value storage |
CA2956216C (en) * | 2014-07-25 | 2020-09-29 | Ab Initio Technology Llc | Mutable chronologies for accommodation of randomly occurring event delays |
WO2016153288A1 (en) * | 2015-03-24 | 2016-09-29 | 건국대학교 산학협력단 | Virtual cluster management system and method for controlling same |
KR101947221B1 (en) * | 2015-03-24 | 2019-02-12 | 건국대학교 산학협력단 | Method and apparatus for scheduling operating time of virtual cluster |
US20160299786A1 (en) * | 2015-04-10 | 2016-10-13 | Microsoft Technology Licensing, Llc | Code examination by scheduler timeline manipulation |
US10592494B2 (en) | 2015-04-14 | 2020-03-17 | Microsoft Technology Licensing, Llc | Collection record for overlapping data stream collections |
US9959137B2 (en) | 2015-04-14 | 2018-05-01 | Microsoft Technology Licensing, Llc | Transaction redo using skip element for object |
US10133768B2 (en) | 2015-04-14 | 2018-11-20 | Microsoft Technology Licensing, Llc | Latest external dependee entity in transaction record |
US10031814B2 (en) * | 2015-04-14 | 2018-07-24 | Microsoft Technology Licensing, Llc | Collection record location as log tail beginning |
US10102251B2 (en) | 2015-04-14 | 2018-10-16 | Microsoft Technology Licensing, Llc | Lockless open collection data structure |
US9766929B2 (en) | 2015-04-14 | 2017-09-19 | Microsoft Technology Licensing, Llc | Processing of data stream collection record sequence |
WO2019070915A1 (en) * | 2017-10-03 | 2019-04-11 | Rubrik, Inc. | Partial database restoration |
US10713124B2 (en) | 2017-11-21 | 2020-07-14 | Rubrik, Inc. | Database snapshot and backup management with recoverable chains |
US10860430B2 (en) * | 2018-07-06 | 2020-12-08 | EMC IP Holding Company LLC | System and method for resilient backup generation |
US10860431B2 (en) * | 2018-07-06 | 2020-12-08 | EMC IP Holding Company LLC | System and method for fault tolerant backup generation in a virtual environment |
US11335141B2 (en) | 2019-10-22 | 2022-05-17 | Argo AI, LLC | Checkpoint-based tracing for monitoring a robotic system |
US11216344B2 (en) | 2020-05-13 | 2022-01-04 | Rubrik, Inc. | Real time database backup status indication and restore |
US11461155B2 (en) * | 2020-05-31 | 2022-10-04 | Wipro Limited | Method and system for predicting an occurrence of a failure condition in a VDI environment |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4665520A (en) * | 1985-02-01 | 1987-05-12 | International Business Machines Corporation | Optimistic recovery in a distributed processing system |
US6078930A (en) * | 1997-02-28 | 2000-06-20 | Oracle Corporation | Multi-node fault-tolerant timestamp generation |
US20030005102A1 (en) * | 2001-06-28 | 2003-01-02 | Russell Lance W. | Migrating recovery modules in a distributed computing environment |
US20070050582A1 (en) * | 2005-08-29 | 2007-03-01 | Searete Llc, A Limited Liability Corporation Of The State Of Delaware | Multi-voltage synchronous systems |
Family Cites Families (27)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5794005A (en) * | 1992-01-21 | 1998-08-11 | The United States Of America As Represented By The Administrator Of The National Aeronautics And Space Administration | Synchronous parallel emulation and discrete event simulation system with self-contained simulation objects and active event objects |
GB2338325B (en) * | 1994-10-03 | 2000-02-09 | Univ Westminster | Data processing method and apparatus for parallel discrete event simulation |
CA2150967C (en) * | 1994-12-22 | 2001-04-03 | Jon C. R. Bennett | Method and a scheduler for controlling when a server provides service with rate control to an entity |
JP3154942B2 (en) * | 1995-09-11 | 2001-04-09 | 株式会社東芝 | Distributed checkpoint generation method and computer system to which the method is applied |
US5917822A (en) * | 1995-11-15 | 1999-06-29 | Xerox Corporation | Method for providing integrated packet services over a shared-media network |
JP3253883B2 (en) * | 1997-01-31 | 2002-02-04 | 株式会社東芝 | Process restart method and process monitoring device |
US6031987A (en) * | 1997-05-06 | 2000-02-29 | At&T | Optimistic distributed simulation based on transitive dependency tracking |
US6397352B1 (en) * | 1999-02-24 | 2002-05-28 | Oracle Corporation | Reliable message propagation in a distributed computer system |
KR100382851B1 (en) * | 1999-03-31 | 2003-05-09 | 인터내셔널 비지네스 머신즈 코포레이션 | A method and apparatus for managing client computers in a distributed data processing system |
US6553515B1 (en) * | 1999-09-10 | 2003-04-22 | Comdial Corporation | System, method and computer program product for diagnostic supervision of internet connections |
US6584581B1 (en) * | 1999-12-06 | 2003-06-24 | Ab Initio Software Corporation | Continuous flow checkpointing data processing |
CA2393740A1 (en) * | 1999-12-08 | 2001-06-14 | The University Of British Columbia | Weighted fair queuing scheduler |
US7315948B1 (en) * | 1999-12-10 | 2008-01-01 | International Business Machines Corporation | Time stamping method employing a separate ticket and stub |
AU5506201A (en) * | 2000-05-04 | 2001-11-12 | Access Co., Ltd. | Information processing method and system |
US6708288B1 (en) * | 2000-10-31 | 2004-03-16 | Hewlett-Packard Development Company, L.P. | Compiler-based checkpointing for support of error recovery |
US7013348B1 (en) * | 2001-12-21 | 2006-03-14 | Nortel Networks Limited | Switch subtending in a synchronous transmission system |
WO2003096235A2 (en) * | 2002-05-13 | 2003-11-20 | Rensselaer Polytechnic Institute | Discrete event simulation system and method |
US7383550B2 (en) * | 2002-12-23 | 2008-06-03 | International Business Machines Corporation | Topology aware grid services scheduler architecture |
US7937551B2 (en) * | 2003-01-21 | 2011-05-03 | Dell Products L.P. | Storage systems having differentiated storage pools |
WO2004099985A1 (en) * | 2003-05-09 | 2004-11-18 | Fujitsu Limited | Execution environment danger prediction/evading method, system, program, and recording medium thereof |
US7886180B2 (en) * | 2004-05-14 | 2011-02-08 | International Business Machines Corporation | Recovery in a distributed stateful publish-subscribe system |
US7873941B2 (en) * | 2004-06-14 | 2011-01-18 | Alcatel-Lucent Usa Inc. | Manager component that causes first software component to obtain information from second software component |
US7660366B2 (en) * | 2004-08-30 | 2010-02-09 | Harmonic Inc. | Message synchronization over a stochastic network |
US20060168338A1 (en) * | 2004-11-23 | 2006-07-27 | Bruegl Aaron R | Methods and systems for providing data across a network |
US7376867B1 (en) * | 2004-12-08 | 2008-05-20 | Hewlett-Packard Development Company, L.P. | Method of seeking consensus among computer processes |
US7903560B2 (en) * | 2004-12-30 | 2011-03-08 | Intel Corporation | Correlation technique for determining relative times of arrival/departure of core input/output packets within a multiple link-based computing system |
EP2163016A2 (en) * | 2007-06-15 | 2010-03-17 | Michele Bargauan | Deterministic program clock reference re-stamping for synchronous bit rate adaptation based on mega-frame initialization packet |
-
2007
- 2007-04-26 US US11/740,556 patent/US7937618B2/en not_active Expired - Fee Related
-
2008
- 2008-04-15 EP EP08742917.1A patent/EP2156307B1/en not_active Not-in-force
- 2008-04-15 AU AU2008244623A patent/AU2008244623B2/en not_active Ceased
- 2008-04-15 CN CN2008800129532A patent/CN101663660B/en not_active Expired - Fee Related
- 2008-04-15 WO PCT/US2008/004866 patent/WO2008133818A1/en active Application Filing
- 2008-04-15 JP JP2010506208A patent/JP5373770B2/en not_active Expired - Fee Related
- 2008-04-15 KR KR1020097016949A patent/KR101126979B1/en not_active IP Right Cessation
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4665520A (en) * | 1985-02-01 | 1987-05-12 | International Business Machines Corporation | Optimistic recovery in a distributed processing system |
US6078930A (en) * | 1997-02-28 | 2000-06-20 | Oracle Corporation | Multi-node fault-tolerant timestamp generation |
US20030005102A1 (en) * | 2001-06-28 | 2003-01-02 | Russell Lance W. | Migrating recovery modules in a distributed computing environment |
US20070050582A1 (en) * | 2005-08-29 | 2007-03-01 | Searete Llc, A Limited Liability Corporation Of The State Of Delaware | Multi-voltage synchronous systems |
Non-Patent Citations (2)
Title |
---|
F. CRISTIAN, F. JAHANIAN: "A timestamp-based checkpointing protocol for long-lived distributed computations", IEEE, 1991 |
S. W. SMITH, D. B. JOHNSON, J. D. TYGAR: "Completely asynchronous optimistic recovery with minimal rollbacks", IEEE, 1995 |
Also Published As
Publication number | Publication date |
---|---|
US20080270838A1 (en) | 2008-10-30 |
EP2156307B1 (en) | 2018-05-23 |
KR20090118934A (en) | 2009-11-18 |
JP5373770B2 (en) | 2013-12-18 |
JP2010525484A (en) | 2010-07-22 |
KR101126979B1 (en) | 2012-03-26 |
AU2008244623B2 (en) | 2013-03-28 |
EP2156307A1 (en) | 2010-02-24 |
US7937618B2 (en) | 2011-05-03 |
AU2008244623A1 (en) | 2008-11-06 |
EP2156307A4 (en) | 2010-10-13 |
CN101663660B (en) | 2012-08-01 |
CN101663660A (en) | 2010-03-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US7937618B2 (en) | Distributed, fault-tolerant and highly available computing system | |
US8893139B2 (en) | System and method for time-aware run-time to guarantee timeliness in component-oriented distributed systems | |
Mishra et al. | Consul: A communication substrate for fault-tolerant distributed programs | |
US6625751B1 (en) | Software fault tolerant computer system | |
US6014673A (en) | Simultaneous use of database and durable store in work flow and process flow systems | |
US7613597B2 (en) | Non-intrusive method for simulation or replay of external events related to an application process, and a system implementing said method | |
US20090313311A1 (en) | Mixed mode synchronous and asynchronous replication system | |
Babaoglu et al. | Enriched view synchrony: A programming paradigm for partitionable asynchronous distributed systems | |
Koldehofe et al. | Rollback-recovery without checkpoints in distributed event processing systems | |
Babaoğlu | Fault-tolerant computing based on Mach | |
Becker | Application-transparent fault tolerance in distributed systems | |
Danilecki et al. | ReServE service: An approach to increase reliability in service oriented systems | |
Takemoto | Fault-tolerant object on network-wide distributed object-oriented systems for future telecommunications applications | |
Dumitraş et al. | Architecting and implementing versatile dependability | |
Goldberg | Optimistic algorithms for distributed transparent process replication | |
Cherif et al. | Replica management for fault-tolerant systems | |
Holenko et al. | The impact of service semantics on the consistent recovery in SOA | |
Koob et al. | Foundations of dependable computing: paradigms for dependable applications | |
Weber et al. | Vidnet: Distributed processing environment for computer generated animation | |
Babaoğlu | Tools and Techniques for Adding Fault Tolerance to Distributed and Parallel Programs | |
Ahn | SBML Protocol for Conquering Simultaneous Failures with Group Dissemination Functionality. | |
Babaoglu | Tools and Techniques for Adding Fault Tolerance to Distributed and Parallel Programs | |
as Bartha | Rollback recovery in distributed systems | |
Babaoğglu et al. | Programming Partition-Aware Network Applications? | |
Kim et al. | A special checkpointing object method based primary-passive TMO replication scheme and middleware support |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
WWE | Wipo information: entry into national phase |
Ref document number: 200880012953.2 Country of ref document: CN |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 08742917 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 1020097016949 Country of ref document: KR |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2010506208 Country of ref document: JP |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2008244623 Country of ref document: AU |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2008742917 Country of ref document: EP |
|
ENP | Entry into the national phase |
Ref document number: 2008244623 Country of ref document: AU Date of ref document: 20080415 Kind code of ref document: A |