PVaniM 2.0-- Online and Postmortem
Visualization Support for PVM

For online visualization analysis, PVaniM 2.0 utilizes sampling to gather data regarding interesting aspects of the application as well as interesting aspects of the cluster environment. With sampling, necessary statistics are collected and sent to the monitor intermittently. The rate at which the application data is sent to the monitor is a parameter that may be set by the user. The user may wish to have the statistics sent to the monitor every 5 seconds, every 30 seconds, once a minute, etc. With lower sampling rates, the application will experience less perturbation, but the graphical views will not be updated as frequently. Currently, PVaniM 2.0 uses 5 seconds as a default sampling rate. Users are encouraged to try different sampling rates and find one that is most suitable to their personal tastes and the needs of their application.

For offline visualization analysis, PVaniM 2.0 utilizes buffered tracing to provide support for fine-grain visualization systems. Currently, PVaniM produces tracefiles for the prototype PVaniM system. A converter is provided that allows the user to also use the popular ParaGraph system using techniques similar to PGPVM.

All of the PVaniM 2.0 system is available via anonymous FTP from ftp.cc.gatech.edu as pub/people/stasko/pvanimOL.tar.Z.

The Host List view lists the hosts utilized and the placement of tasks on each of these hosts. The tasks are represented by their index in a user-defined task array passed into the system by the application.

The Load Information view provides insight into the load on the host machines by providing a graphical view of the average number of jobs in the run queue of the host.

In the Host Utilization view, a rectangle is drawn composed of green, yellow, and red portions. Green is used to represent the amount of time the host spends computing, yellow represents the amount of time the host performs PVM message sends, and red represents the amount of time the host is idle waiting to receive PVM messages. Note that for each host, utilization is a composition of the PVM tasks running on that host.

This Task Summary view provides the same information as the Host Utilization view but provides this information on a per task basis.

The Memory Use view illustrates the aggregate amount of memory utilized by the PVM tasks on each host. By clicking a mouse button on rectangles in the Memory Use view, more detailed textual usage statistics will be placed in olPVaniM's output box. The same information is also available by clicking the rectangles in either the Load Information or Host Utilization views.

The MESSAGES SENT/bytes sent view provides information regarding both the number of messages sent and the number of bytes sent. By clicking with the mouse on the ``MESSAGES SENT/bytes sent'' label, the user may interactively choose the type of information presented by this view. It is worth mentioning that this view does not present aggregate statistics but instead only interval statistics. That is, the view illustrates how much message traffic has occurred by a task since the last time it reported statistics to the monitor.

The TOTAL MATRIX COMM/interval matrix comm view provides both aggregate and interval statistics regarding message communication. In this view, a row of squares is provided for each task where each square represents how much communication occurs between that task and the task associated with the square. In its default form, this view provides aggregate statistics and the message key scales as the number of messages increases. By clicking with the mouse on the ``TOTAL MATRIX COMM/interval matrix comm'' label, the view switches to providing this information on an interval basis. That is, the squares are colored using only messages sent since the last time the task reported to the monitor. When used in this mode, it is not clear what the range values should be for the message legend. Because of this, the user is allowed to interactively change the range values of the message legend to find a range most appropriate for the phase of the application. By clicking the right or middle mouse button, the range values increase. By clicking the left mouse button, the range values decrease.