Grabner, S., Kranzlmller, D., Volkert, J., Debugging parallel programs using ATEMPT, Proceedings of HPCN Europe 95 Conference, Milano, Italy, May, 1995.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....transactions rules and the objects involved by a debugger. Furthermore, techniques for debugging parallel programs should be adopted. For instance, the trace driven event replay technique as already supported by TriGS Debugger could be used for semi automatic detection of race conditions [Grab95]. A History of Database States would Ease Debugging. Trace data gathered by TriGS Debugger represents the history of event detections and rule executions, but it does not include the history of related database objects. This leads to a problem already discussed in Section 3.2, namely that one can ....

Grabner, S., Kranzlmller, D., Volkert, J., Debugging parallel programs using ATEMPT, Proceedings of HPCN Europe 95 Conference, Milano, Italy, May, 1995.

....were published ( 2] 18] 21] 24] 4] but no comprehensive approach summing up state of the art techniques to a common on line monitoring system has been proposed. In order to avoid the complexity of on line monitoring a lot of tools are based on traces. This refers to debugging tools [10] as well as to performance analysis [13] A main reason for the popularity of traces is the existence of several standard formats, like SDDF [23] or PICL [9] Already in 1991 B. Bruegge was proposing a heterogeneous on line monitor, called BEE (basis for distributed event environments) 2] albeit ....

S. Grabner, D. Kranzlm uller, and J. Volkert, Debugging parallel programs using ATEMPT, in High Performance Computing and Networking 1995, B. Hertzberger and G. Serazzi, eds., no. 919 in Lecture Notes in Computer Science, Springer, May 1995, pp. 235--240.

....A decision was needed on whether an integrated tools environment was desirable. The term integrated can be used in the sense that all tools have a similar feel. Integration can also imply that the output from one tool feeds into another, a toolset as opposed to a toolkit. In the MAD environment [ 10 ] there are: EMU, which instruments and monitors the application; ATEMPT, which provides performance analysis and error detection through visualization; and 4 PARASIT, which simulates the application in order to predict race conditions. We decided to avoid the attempt to lock the user into a ....

....one task in the dictionary stage were set with the same time. Outward message sizes were 2119 bytes in stage one, less than 15 bytes in stage two, and 112 bytes maximum in stage three. Interstage buffering had been set by trial and error at 20 slots, while the local input buffer sizes were 10 slots. The aim had been to find the best throughput if jobs were instantaneously available in the worst case scenario. Each application has some special features. In the postcode application, differing postcode (task) sizes in the final stage occur which we bracketed by worst (all size seven) and ....

S. Grabner, D. Kranzlmuller, and J. Volkert. Debugging parallel programs using ATEMPT. In B. Hertzberger and G. Serazzi, editors, High-Performance Computing and 18 Networking International Conference, pages 235--240. Springer, Berlin, 1995. Lecture Notes in Computer Science Volume 919.

....a record with minimal perturbation. Generally, one should bear in mind that the pattern of message passing on a distributed system might be different to that of the target machine. The intention is to catch unexpected orderings. A scalar logical clock also is employed in the ATEMPT trace system [ 55 ] At this stage in development, the trace display was via the post mortem visualizer, ParaGraph [ 56 ] A standard trace file [ 57 ] format was used, compatible with ParaGraph. The format includes a broadcast field but does not include multicast, which is understandable as the destinations are ....

S. Grabner, D. Kranzlmuller, and J. Volkert. Debugging parallel programs using ATEMPT. In B. Hertzberger and G. Serazzi, editors, High-Performance Computing and Networking International Conference, pages 235--240. Springer, Berlin, 1995. Lecture Notes in Computer Science Volume 919.

....were published ( 1] 10] 11] 13] 2] but no comprehensive approach summing up state of the art techniques to a common on line monitoring system has been proposed. In order to avoid the complexity of on line monitoring many tools are based on traces. This refers to debugging tools [5] as well as to performance analysis [6] A main reason for the popularity of traces is the existence of several standard formats, like SDDF [12] or PICL [4] In addition, the development of tools and monitor systems is almost never done by independent research groups. As a result, tools run only ....

S. Grabner, D. Kranzlmuller, and J. Volkert. Debugging parallel programs using atempt. In B. Hertzberger and G. Serazzi, editors, High Performance Computing and Networking 1995, number 919 in Lecture Notes in Computer Science, pages 235--240. Springer, May 1995.

....were published ( 1] 10] 11] 13] 2] but no comprehensive approach summing up state of the art techniques to a common on line monitoring system has been proposed. In order to avoid the complexity of on line monitoring many tools are based on traces. This refers to debugging tools [5] as well as to performance analysis [6] A main reason for the popularity of traces is the existence of several standard formats, like SDDF [12] or PICL [4] The goal of the OMIS (on line monitoring interface specification) project is to separate tool and monitor development by providing a ....

S. Grabner, D. Kranzlmuller, and J. Volkert. Debugging parallel programs using atempt. In B. Hertzberger and G. Serazzi, editors, High Performance Computing and Networking 1995, number 919 in Lecture Notes in Computer Science, pages 235--240. Springer, May 1995.

....each time an event takes place. These are all the data which are necessary for the post mortem analysis of the program run. 2. 2 ATEMPT A Tool for Event Manipulation The tracefiles of EMU are read by ATEMPT, A Tool for Event ManiPulaTion, which visualizes them as a global communication graph [Grab 95a] This eventgraph consists of vertices which define the occurrence of an event and of arcs which define the sequential part between the events or the communication between corresponding send and receive events. As it is not possible to determine the correct occurrence time of an event due to the ....

S. Grabner, D. Kranzlmüller, J. Volkert, "Debugging parallel programs using ATEMPT", Proc. of HPCN Europe 95 Conference, Milano, Italy (May 1995).

No context found.

S. Grabner, D. Kranzlmüller, J. Volkert, "Debugging Parallel Programs using ATEMPT", Proc. of HPCN Europe 95, Milan, Italy (May 1995)

....non determinism in the program flow, which can lead to very serious and sporadically appearing faults. With ATEMPT it is possible to detect events which are possible candidates for race conditions. The verification of these race condition candidates is done with the trace driven simulator PARASIT [9] which tries to filter race conditions that actually produce faults. 5. Performance Evaluation. Another part of ATEMPT covers the field performance evaluation. It was established during the design phase of ATEMPT because most of the data was already available. The difference between error ....

S. Grabner, D. Kranzlmüller, J. Volkert, "Debugging parallel programs using ATEMPT", submitted to: HPCN Europe 95 Conference, Milano, Italy (1995).

No context found.

S. Grabner, D. Kranzlmller, J. Volkert, "Debugging Parallel Programs using ATEMPT", Proc. of HPCN Europe 95, Milan, Italy (May 1995)

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