Hollingsworth, J. K. Finding Bottlenecks in Large-Scale Parallel Programs. PhD Thesis, University of Wisconsin-Madison, August 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....with detailed knowledge about the architecture to detect performance bottlenecks. Such a system should be able to capture data from all relevant sources while incurring minimal overhead on the application. It should be flexible enough to support advanced data capturing and data analysis techniques [10, 8]. Finally, it should be scalable up to the maximum size Origin system. The goal of this work is to design and implement a low overhead, tracebased scalable performance data gathering system. The evaluation work will investigate the total overhead and impact on the application and the tradeoffs in ....

....and presented in the source browser window. SvPablo supports Portland Group HPF compiler trace mechanism as well. Paradyn performance instrumentation and visualization tool supports realtime insertion and removal of instrumentation code during program execution. A technique called W search [8] is used to automatically search for performance bottlenecks. The system uses real time instrumentation to test hierarchical tree of progressively refined hypotheses describing possible causes for program bottlenecks. Digital continuous profiling infrastructure [2] developed a system for ....

Hollingsworth, J. K. Finding Bottlenecks in Large-Scale Parallel Programs. PhD Thesis, University of Wisconsin-Madison, August 1994.

....captured at the network or switch levels[5, 23] At this time, we are implementing network monitoring by inspection of kernel resident protocol stacks using the kernelresident libpcap portion of the well known tcpdump facility. Higher level services that interpret or analyze monitoring data[15, 21] are not the subject of this research, but would be useful when using monitoring data for runtime program steering[22] or to help programmers tune their cluster applications[21] Dproc uses the open source nature of the Linux kernel and its ability to dynamically link new modules into the kernel. ....

Jerey K. Hollingsworth. Finding Bottlenecks in Large-scale Parallel Programs. Ph.D. Dissertation, August 1994. 11-127, Geneva, Switzerland, April 1997

....captured at the network or switch levels[6, 28] At this time, we are implementing network monitoring by inspection of kernel resident protocol stacks using the kernelresident libpcap portion of the well known tcpdump facility. Higher level services that interpret or analyze monitoring data[16, 24] are not the subject of this research, but would be useful when using monitoring data for runtime program steering[25] or to help programmers tune their cluster applications[24] Dproc uses the open source nature of the Linux kernel and its ability to dynamically link new modules into the kernel, ....

Jerey K. Hollingsworth. Finding Bottlenecks in Large-scale Parallel Programs. Ph.D. Dissertation, August

....successively detailed regions of the program is provided, one can proceed in a hierarchical manner with respect to the program code, by studying the code at various levels of detail. This helps to systematically focus on the region of code causing the performance problem. A previous informal study [10] shows that users generally start by taking a high level view of the performance of their applications, and isolate the source of their program s poor performance. This process continues until they understand their program s performance well enough to start tuning it. Therefore, the fourth goal of ....

J. K. Hollingsworth, Finding Bottlenecks In Large Scale Parallel Programs, PhD Thesis, University of Wisconsin-Madison, 1994.

....how we solved a set of particular run time performance problems where a foreign agent (such as a competing process, inadequate memory, or a full disk) on one or more machines slowed down the entire application. In general, this was a hierarchical process, similar to the search process described in [15]. In the first step, finding the needle in a NOW stack , the slow workstations in the cluster were identified. In the second step, removing the needle , the specific problem on those slow workstations was identified and solved. 6.1 Finding the Needle Because each process in our parallel sort ....

J. K. Hollingsworth. Finding Bottlenecks in Large-scale Parallel Programs. PhD thesis, University of Wisconsin, Aug. 1994.

No context found.

Hollingsworth 94 J. K. Hollingsworth. Finding Bottlenecks in Large Scale Parallel Programs. PhD Thesis, Dept. of Computer Science, University of Wisconsin-Madison, 1994.

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