J. Hollingsworth and P. Keleher. Prediction and adaptation in Active Harmony. In Proc. 7th Intl. Symp. on High Performance Distributed Computing, Apr. 1998. 142  Home/Search   Document Details and Download   Summary   Related Articles   Check

....where the network resources should be treated as a given, the application can itself be adapted to achieve desired QoS requirements. The techniques that have been proposed can be further broken down into two classes. The first class comprises systems such as EPIQ [7] ErDos [6] Active Harmony [14], and our own Application Tunability framework [4] which assume that the application structure is more or less fixed and adaptation is achieved by altering the internal behavior of one or more of the components (e.g. changing an internal algorithm) The second class of approaches, exemplified ....

J. K. Hollingsworth and P. J. Keleher. Prediction and adaptation in active harmony. Cluster Computing, 2(3):195--205, 1999.

....of responses to assertions that a user can select using the PA definition or an environment variable. The action can be ignored, recorded to a log, trigger more detailed monitoring, invoke a user defined callback, or activate some corrective action, possibly using an adaptation system like Harmony [7] or Autopilot [13] PA actions are, by default, counters that accumulate the number of failures for an assertion. To specify one of alternative actions mentioned above, users simply call the pa set action subroutine with the appropriate parameters after defining the assertion. Subsequent ....

J.K. Hollingsworth and P. Keleher, "Prediction and Adaptation in Active Harmony," Proc. HPDC, 1998, pp. 180-8.

....in an important property of the SFO paradigm. Adaptation is necessary when the network resources fluctuate in performance such as in a dynamic network environment. A number of research groups are exploring the concept of application adaptivity including Autopilot [11] and Active Harmony [8]. Autopilot is a toolkit that contains a library of runtime components needed to build adaptive applications. Active Harmony is exploring the concept of adaptability for threaded parallel applications. Madhyastha and Reed 5 have investigated adaptive file systems in which the I O system learns and ....

J.K. Hollingsworth and P.J. Keleher, "Prediction and Adaptation in Active Harmony," Proceedings of the Seventh IEEE International Symposium on High Performance Distributed Computing, July 1998.

....and recursive types were not supported, all shared data had to be accessed indirectly through a local mapping table, and only a single memory model (similar to processor consistency) was supported. Perhaps the two projects most closely related to InterWeave are Khazana [9] and Active Harmony [21]. Both are outgrowths of previous work in software distributed shared memory. Both support distributed sharing without enforcing an object oriented programming style. Khazana proposes a global, 128 bit address space for all the world s shared data. It does not impose any structure on that data, or ....

J. K. Hollingsworth and P. J. Keleher. Prediction and Adaptation in Active Harmony. In Proc. of the 7th Intl. Symp. on High Performance Distributed Computing, Chicago, IL, Apr. 1998.

....and recursive types were not supported, all shared data had to be accessed indirectly through a local mapping table, and only a single memory model (similar to processor consistency) was supported. Perhaps the two projects most closely related to InterWeave are Khazana [9] and Active Harmony [21]. Both are outgrowths of previous work in software distributed shared memory. Both support distributed sharing without enforcing an object oriented programming style. Khazana proposes a global, 128 bit address space for all the world s shared data. It does not impose any structure on that data, or ....

J. K. Hollingsworth and P. J. Keleher. Prediction and Adaptation in Active Harmony. In PROC of the 7TH HPDC, Chicago, IL, APR 1998.

.... analysis, and code or environment adjustments to improve performance[66] There is also research into self tuning applications that include code to trigger changes based on measured performance values, either as a prelude to a full execution [75] or online as the application is running [31]. The tools detailed above address single iterations of one or more steps in the tuning process. Our approach is to provide a framework that can incorporate existing techniques for each step of the performance tuning cycle, and extend the scope to include repeated iterations. Hondroudakis and ....

J.K. Hollingsworth and P.J. Keleher. Prediction and adaptation in Active Harmony. Proceedings of the 7th International Symposium on High Performance Distributed Computing, pages 180--188, Chicago, IL, 1998.

....PerDiS and a few CORBA systems (e.g. Fresco [22] cache objects for locality of reference. Thor [25] enforces type safe object oriented access to records in a heterogeneous distributed database. Perhaps the two projects most closely related to InterWeave are Khazana [7] and Active Harmony [18]. Both are outgrowths of previous work in software distributed shared memory. Both support distributed sharing without enforcing an object oriented programming style. Khazana proposes a global, 128 bit address space for all the world s shared data. It does not impose any 4 structure on that data, ....

J. K. Hollingsworth and P. J. Keleher. Prediction and Adaptation in Active Harmony. In Proc. of the 7th Intl. Symp. on High Performance Distributed Computing, Chicago, IL, Apr. 1998.

....robust split and merge operations, decision algorithms, monitoring support, etc. ACDS controlled computational data streams may be used with arbitrary parallel applications, including those written with meta computing systems like Globus [FK97] Legion [GW96] Schooner [HS94] or Active Harmony [HK98] In these contexts, ACDS addresses only the runtime control of the computational data streams linking such Grid computations to end users. In comparison, the load balancing and resource management mechanisms included with the grid computing frameworks themselves concern the runtime managent of ....

J. K. Hollingsworth and P. J. Keleher. Prediction and adaptation in active harmony. In Proceedings of the Seventh IEEE International Symposium on High Performance Distributed Computing, 1998.

....consistency. The support of release consistency allows Proteus to delay propagation of modifications to shared memory, thereby shortening the memory access latency. The third is that Proteus tries to balance workloads among nodes when workload redistribution is performed. The Active Harmony[10] attempts to provide a global resource management which can accommodate to changes of a node configuration. While this system supports DSM among nodes, its goal is focused on the prediction of when to migrate applications and whether to migrate the process or the data. Proteus, on the other hand, ....

Hollingsworth, Jeffrey K. andKeleher, Peter J., Prediction and Adaptation in Active Harmony, In HPDC '98.

....in an important property of the SFO paradigm. Adaptation is necessary when the network resources fluctuate in performance such as in a dynamic network environment. A number of research groups are exploring the concept of application adaptivity including Autopilot [14] and Active Harmony [10]. Autopilot is a toolkit that contains a library of runtime components needed to build adaptive applications. Active Harmony is exploring the concept of adaptability for threaded parallel applications. The SFO employs adaptivity that is focused on the problem of remote I O. 3.0 The Concept The ....

J.K. Hollingsworth and P.J. Keleher, "Prediction and Adaptation in Active Harmony," Proceedings of the Seventh IEEE International Symposium on High Performance Distributed Computing, July 1998.

....is relatively small. For many programs we have harmonized the change amounted to less than 50 lines of code. The full paper includes an example of specifying application and library parameters using the Harmony API. 5. Parameter Tuning Algorithm In an earlier version of the Harmony system [7], we had a simple greedy algorithm to handle automatic selection of the appropriate parameters. However, for larger applications a more sophisticated alga rithm is needed. The problem of selecting good parameters reduces to finding a k tuple in the value space determined by the values of the ....

....for applications to disclose their specific preferences, Harmony will encourage programmers to think about their needs in terms of options and their characteristics rather than as selecting from specific resource alternatives described by the system. Prior work in the active Harmony project [7][8] concentrated on the API to make applications tunable, and in defining an interface to express the different options via a Resource Specification Language. This paper extends that work be providing an improved search algorithm (rather than a simple greedy approach) In addition, we describe ....

J. K. Hollingsworth and P. J. Keleher, "Prediction and Adaptation in Active Harmony," Cluster Computing, 2 (1999), pp. 195-205

....of critical path analysis for parallel programs running on SMPs[6] In this application, instrumentation code to track the synchronization events in a shared memory program is inserted on demand using the API. At Maryland, we are also investigating using Dyninst as part of the Harmony project[7]. The Harmony project is looking into using runtime observations of applications to automatically tune programs by selecting from candidate configurations. Runtime code patching will be used to change which versions of procedures and libraries get called at specific locations. A similar use of the ....

J. K. Hollingsworth and P. J. Keleher, "Prediction and Adaptation in Active Harmony," The 7th International Symposium on High Performance Distributed Computing. July 1998, Chicago, pp. 180-188.

No context found.

J. Hollingsworth and P. Keleher. Prediction and adaptation in Active Harmony. In Proc. 7th Intl. Symp. on High Performance Distributed Computing, Apr. 1998. 142

No context found.

J. K. Hollingsworth and P. J. Keleher, "Prediction and Adaptation in Active harmony", The Seventh International Symposium on High Performance Distributed Computing, July 1998, pg. 180-188.

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