Kamath M., Alonso G., Guenthoer R., and Mohan C. Providing high availability in very large workflow management systems. Technical report, IBM Almaden Research Center, July 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....tasks specifications, and scheduling approaches. Workflow reliability and recovery is still a new area. Since the architecture of the workflow management system is distributed, managing local system failures and communication failures is of utmost importance. Some work can be found in [2, 17, 19, 22] and in the Exotica Project [1, 20, 21] Empirical studies of workflow system architectures are scarce. These studies are needed to evaluate the different systems and to verify the benefits of using the workflow technology in real organizations. It is also important in the direction of ....

M. Kamath, G. Alonso, R. Gunthor, and C. Mohan. Providing High Availability in Very Large Workflow Management Systems. In Proc. 5th International Conference on Extending Database Technology, Avignon, March 1996.

....Role Coordinator. More complex queries, requesting history information could increase the negotiation phase of the agent in some seconds. This query, however, is specific of some activities and is out of the scope of the present work. 5. Related Work Some of the components of the Exotica project [9], 10] 11] 12] developed at IBM Almaden Research Center, have similarities to our proposal. In particular the Exotica FMQM (Flowmark on Message Queue Manager) architecture is a distributed model for WFMSs, using a proprietary standard (MQI Message Queue Interface) of persistent queues. The ....

Kamath, M. and Alonso, G. and Gunthor, R. and Mohan, C. 1996. Providing High Availability in Very Large Workflow Management Systems, from Proceedings of the Fifth International Conference on Extending Database Technology (EDBT'96). Avignon, France. Pages 2529.

.... for many years in specific TP environments with a small number of carefully administered servers [7, 57] but applying and generalizing these techniques to an open, manifold world of data servers, workflow servers, and business object application servers is still an issue that is all but trivial [44]. In 22 particular, it is unclear to what extent the various replication techniques for data as well as processes scale up with the ever increasing system size and complexity. Like with reliability, one should ideally be able to quantify the availability that can be guaranteed with a given system ....

Kamath, M., Alonso, G., Gunthor, R., Mohan, C.: Providing High Availability in Very Large Workflow Management Systems, Proceedings of the 5th International Conference on Extending Database Technology (EDBT), Avignon, France, 1996.

....systems increases with the number of applications being served, the number of workflow applications and WFMS users within a company will significantly grow year by year once it has started to go that way. Thus, the question arises how to manage large numbers of users (may be even tens of thousands [9]) and high volume data transfer (e.g. in conjunction with multi media applications) within a WFMS. Most existing systems use a central WF server. It is easy to see that it becomes a bottleneck and will be overloaded under a high load. To reduce the load of the WF server it can be replicated. ....

M. Kamath, G. Alonso, R. Gunthor, and C. Mohan. Providing High Availability in Very Large Workflow Management Systems. In Proc. of the 5th Int. Conf. on Extending Database Technology, pages 427--442, Avignon, Mar. 1996.

....methods on the network and workflow server load, mostly using simulations. 28] presents heuristics for the allocation of workflow type and workflow instance data onto servers. Mechanisms for enhanced WFMS availability by replicating state data on a standby backup server have been studied in [11, 17]. 16] characterizes the workload of cross organizational workflows by means of Markov models. None of this prior work has addressed the issue of how to systematically benchmark a WFMS and how to configure a WFMS for given performance goals. Benchmarking is well established, with benchmark ....

M. Kamath, G. Alonso, R. Gnthr, C. Mohan, Providing High Availability in Very Large Workflow Management Systems, 5 th Int'l Conf. on Extending Database Technology (EDBT), Avignon, France, 1996

....architecture objects are written in Java, their persistence are implemented using the java.io.Serializable interface. All the objects managed by the LOA are periodically serialized and stored in their respective Object Repositories. 5. Related Work Some of the components of the Exotica project [5,6,7,8], developed at IBM Almaden Research Center, have similarities to our proposal. In particular the Exotica FMQM (Flowmark on Message Queue Manager) architecture is a distributed model for workflows, using a proprietary standard (MQI Message Queue Interface) of persistence queues. The case is a ....

Kamath, G. Alonso, R. Gnthr, C. Mohan. "Providing High Availability in Very Large Workflow Management Systems", In Proceedings of the Fith International Conference on Extending Database Technology (EDBT'96), Avignon, France, march 2529, 1996.

....methods on the network and workflow server load, mostly using simulations. 18] presents heuristics for the allocation of workflow type and workflow instance data onto servers. Mechanisms for enhanced WFMS availability by replicating state data on a standby backup server have been studied in [9, 14]. None of this prior work has addressed the issue of how to configure a WFMS for given performance and availability goals. The use of CTMC models in the context of workflow management has been pursued by [13] This work uses the steady state analysis of such models to analyze the efficiency of ....

M. Kamath, G. Alonso, R. Gnthr, C. Mohan, Providing High Availability in Very Large Workflow Management Systems, Int'l Conf. on Extending Database Technology (EDBT), Avignon, France, 1996

....concern in this project. Other notable research projects with similar objectives are the DOM project [GH94] which is primarily oriented towards applications in the telecommunications industry and views workflows as a set of control flow dependencies between transactional steps, the Exotica project [MAGK95, KAGM96] which aims to enhance the FlowMark product by addressing scalability and availability issues, and the METEOR project [KS95] as well as the MOBILE project [Ja94] both of which pursue workflow execution models driven by intertask dependency rules that are expressed in specifically designed script ....

M. Kamath, G. Alonso, R. Gunthor, C. Mohan, Providing High Availability in Very Large Workflow Management Systems, International Conference on Data Engineering, 1996

.... loans only to credit worthy customers, or purchasing a gameboy within certain financial constraints) Thus, this kind of service quality requires formal reasoning on the specification of the workflow [8] but also failure resilience and availability guarantees from an underlying recovery manager [9, 10]. ffl On the quantitative side, consider the notion of quality of service in the area of multimedia storage and communication [11, 12] This involves quantifying the presentation quality of video and audio streams in terms of guaranteed bandwidth or guaranteed playback smoothness . In addition, ....

Kamath, M., Alonso, G., Gunthor, R., Mohan, C.: Providing High Availability in Very Large Workflow Management Systems, Proceedings of the 5th International Conference on Extending Database Technology (EDBT), Avignon, France, 1996.

....does not require a centralized context database. Process instances can be distributed among multiple context databases. In order to improve performance and availability the persistent image of a process can be replicated by the context database using the techniques that are discussed in [KGM96, PG94]. Persistent context of a process instance can be accessed by different workflow engines but only one workflow engine can access at a time. The process context holds three kinds of information: 1) The process structure and the current state; 2) the workflow relevant data (WRD) and (3) the ....

....entities. The ORB can be a performance bottleneck in very large workflow applications. However, the application may be divided into manageable parts that are running on different ORBs. CORBA hides the ORB boundaries from the rest of the system. Availability is another important concern for WFMSs [KGM96]. Dflow can incorporate well known database replication techniques in its storage wrapper. Different availability levels can be enforced for each persistent object instance according to the system configuration. A backup server can be used against hardware failures. CORBA allows multiple servers ....

M. Kamath, G. Alonso, R. Gunthor, C. Mohan, "Providing High Availability in Very Large Workflow Management Systems", International Conference on Extending Database Technology, 1996.

....the execution of processes even in the event of failures. Such requirements define a wide range of failure handling capabilities that the system must support to be commercially viable. Part of our research has been to analyze the possible failure scenarios and design methods to handle them [1, 2, 17]. In the first place, the execution of an activity within a process involves several components: the database server where the workflow process related data is stored, the workflow server that determines where the activity 19 is going to be executed, the client interacting with the user, and the ....

....incorporation and exclusion of new servers without interrupting the system, message duplication, and so forth. We have addressed all these issues and proposed a replication mechanism for workflow management systems that greatly improves the availability of the system avoiding excessive overhead [17]. 3.4 Mobile Computing Workflow management systems must coordinate users distributed over a wide geographic area. The basic idea being that the processes to be executed are defined and controlled in a centralized server, while the users can perform parts of these processes at remote clients. ....

M. Kamath, G. Alonso, R. Gunthor, C. Mohan, Providing High Availability in Very Large Workflow Management Systems, Research Report, IBM Almaden Research Center, in preparation, January 1995.

....In practice, since the PSS is built on top of a database, the backup mechanisms can either be based on the replication capabilities of the underlying DBMS, or implemented at the application level using semantic knowledge of the application. The latter has been suggested as the best approach [KGAM96] due to its added flexibility but, so far, no effort has been made to evaluate these different strategies and check their applicability in a real setting. This paper presents a comprehensive study of backup mechanisms in the context of process support systems. Several algorithms are proposed and ....

....at a high level, without having to rely on the database idiosyncrasies. This approach has significant advantages. First, it allows to use different databases as primary and backup (in the current implementation, Oracle and ObjectStore are both used as backup for each other) which was suggested in [KGAM96] as one of the advantages of semantic backup, although without any empirical evidence. To some degree this argument is losing strength, given that there are now products on the market that allow to replicate data between heterogeneous DBMS [Kno97, IBM97] These products do not yet provide, ....

[Article contains additional citation context not shown here]

M. Kamath, R. Gunthor, G. Alonso, and C. Mohan. Providing high availability in very large workflow management systems. In Proc. of the Fifth International Conference on Extending Database Technology (EDBT), Avignon, France, March 1996.

....Foundation under the project TRAMs (Transactions and Active Database Mechanisms for Workflow Management) replication capabilities of the underlying DBMS, or implemented at the application level using semantic knowledge of the application. The latter has been suggested as the best approach [KGAM96] due to its added flexibility but, so far, no effort has been made to evaluate these different strategies and check their applicability in a real setting. This paper presents a comprehensive study of backup mechanisms in the context of process support systems. Several algorithms are proposed and ....

....a high level, without having to rely on the database idiosyncrasies. This approach has significant advantages. First, it allows to use different databases as primary and backup (in the current implementation, Oracle and ObjectStore are both used as backup for each other) which was suggested in [KGAM96] as one of the advantages of semantic backup, although without any empirical evidence. To some degree this argument is losing strength, given that there are now products on the market that allow to replicate data between heterogeneous DBMS [Kno97, IBM97] These products do not yet provide, ....

[Article contains additional citation context not shown here]

M. Kamath, R. Gunthor, G. Alonso, and C. Mohan. Providing high availability in very large workflow management systems. In Proc. of the Fifth International Conferenceon Extending Database Technology(EDBT), Avignon, France, March 1996.

....the execution of processes even in the event of failures. Such requirements define a wide range of failure handling capabilities that the system must support to be commercially viable. Part of our research has been to analyze the possible failure scenarios and design methods to handle them [1, 2, 18]. In the first place, the execution of an activity within a process involves several components: the database server where the workflow process related data is stored, the workflow server that determines where the activity is going to be executed, the client interacting with the user, and the ....

....to allow other components to reroute their operations to components that are still available. This implies using multiple connections among components. We have designed a new architecture for WFMSs in which the notion of clusters of servers is used to provide enhanced availability of the system [2, 18]. By using several clusters, each of them attached to a separate database, the impact of failures on overall performance and availability can be reduced. Since each cluster s database contains the same schema information such as process templates, role and staff assignments, and organization ....

[Article contains additional citation context not shown here]

M. Kamath, G. Alonso, R. Gunthor, C. Mohan, Providing High Availability in Very Large Workflow Management Systems, Research Report RJ9967, IBM Almaden Research Center, July 1995.

.... are large corporations in which the number of potential users can be in the tens of thousands, the number of concurrent process in the hundreds of thousands, and the number of sites connected to the WFMS in the thousands, distributed over a wide geographic area and based on heterogeneous systems [34]. In such environments, availability is a key feature. Fortunately, most failures in a workflow system can be masked using the redundancy inherent to the architecture. For instance, it is common to have the same application installed in several nodes. If one of them is not available, it may be ....

.... One way to address these concerns is to provide a backup architecture that is database independent, uses knowledge of the semantics of workflow operations to optimize the exchange of information between the primary and the backup, and allows to adjust the degree of availability in the system [34]. For this purpose, standard database techniques such as hot standby, cold standby, 1safe, and 2 safe, can be used [26] These approaches can be combined to provide a flexible mechanism for high availability on workflow systems. Three process categories are defined: normal, important and critical. ....

M. Kamath, G. Alonso, R. Gunthor, C. Mohan. "Providing High Availability in Very Large Workflow Management Systems", Proceedings of the Fifth International Conference on Extending Database Technology (EDBT96), Avignon, France, March 25-29, 1996.

....abstraction layer implements the mechanisms necessary to make the system database independent. The experience with workflow systems shows that this is a crucial issue affecting scalability and the overall openness of the system [AAEM97] Hence, OPERA uses internally a canonical representation [KAGM96] This canonical representation is not suitable, however, for either commercial databases or user interaction. Thus, the database abstraction layer translates the canonical representation to the private representations of the underlying repositories (SQL, C , system calls) as required by the ....

....when performance and scalability is considered. This also opens up the possibility for OPERA to perform automatic load balancing by spawning new navigators and new instance spaces at other sites as the load increases. Several OPERA systems can also be interconnected to form a larger system. As in [KAGM96] the user can chose among three levels of availability for each process. The highest level of availability (critical) guarantees a hotstandby, 2 safe backup. If the primary system fails, the backup can take over immediately. The intermediate level of availability (important) guarantees a ....

M. Kamath, G. Alonso, R. Gunthor, and C. Mohan. Providing High Availability in Very Large Workflow Management Systems. In In Proceedings of EDBT'96 , Avignon, France, March 1996. SpringerVerlag, LNCS 1057.

No context found.

Kamath M., Alonso G., Guenthoer R., and Mohan C. Providing high availability in very large workflow management systems. Technical report, IBM Almaden Research Center, July 1995.

No context found.

M. Kamath, G. Alonso, R. Guenthor, and C. Mohan. Providing high availability in very large workflow management systems. In Proc. of 5th Int. Conf. on Extending Database Technology, 1996.

No context found.

M. Kamath, G. Alonso, R. Gnthr, C. Mohan, Providing High Availability in Very Large Workflow Management Systems, 5 th Int'l Conf. on Extending Database Technology (EDBT), Avignon, France, 1996

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC