"Visualizing the performance of parallel programs" -- Michael T. Heath and Jennifer A. Etheridge

The main design goals of ParaGraph are:

• Easy to understand
  It aims at making visual displays as intuitively meaningful as possible. More than 20 displays or views based on the same underlying trace data are provided.
• Easy to use
  Color and animation are used to make ParaGraph entertaining to use. It has an interactive, mouse- and menu-oriented user interface so its features are easily invoked and customized. ParaGraph's input are trace files produced by Portable Instrumented Communication Library (PICL), which lets users produce trace data automatically.
• Portability
  First, the graphics itseld should be portable. Second, the package must be able to display execution behavior from different parallel architectures and parallel-programming paradigms. ParaGraph inherits a high degree of portability from PICL.

ParaGraph relies on PICL to deal with problems related to clock synchronization; PICL goes to considerable lengths to synchronize processor clocks and adjust for potential clock drift, so timestamps are as meaningful and consistent as possible. PICL also tries to minimize tracing perturbation by saving trace data in each processors local memory, downloading them to disk only after the program has finished execution.

Displays

• Utilization
  Some of the displays that fall into this category are: Processor count, Gantt chart, Summary, Concurrency profile, Utilization meter, and Kiviat diagram.

  The utilization-count display shows the total number of procesors in each of three states - busy, overhead, and idle - as a function of time. Gantt chard depicts the activity of individual procesors with a horizontal bar chart in which the color of each bar indicates busy/ovehead/idle status of the corresponding processor as a function of time. The utilization-summary display shows the percent of time over the entire run that each processor spent in each of the three states (busy, overhead, and idle). The concurrency profile shows the percentage of time that a certain number of processors were in a given state. The utilization-meter indicates the percentage of the maximum number of processors that are currently in each of the three states. The Kiviat diagram gives a geometric depiction of individual processor's usage and the overall load balance across processors.

• Communication
  Some of the displays that fall into this category are: Message Queues, Communication matrix, Animation, Hypercube, Communication meter, Communication traffic, and Space-time diagram.

  The Message Queue display depicts the size of each processor's incoming-message queue with a vertical bar whose height varies with time as messages are sent, buffered, and received. In the communication matrix display, messages are represented by squares in a two-dimensional array whose rows and columns correspond to each message's sending and receiving processor. In the Animation display, the multiprocessor is represented by a graph whose nodes represent processors, and whose arcs represent communication links. The Hypercube display is similar to Animation except that it provides additional node layouts to exhibit more clearly communication patterns that correspond to the various networks that can be embedded in a Hypercube. The Communication meter uses a vertical bar to indicate the current communication volume as a percentage of the maximum. The communication traffic is a simple plot of the total communication traffic in the interconnection network as a function of time. The space-time diagram depicts interactions among processors through space and time.

• Task Information
  Some of the displays that fall into this category are: Task Gantt, Task summary, among others.

  Task Gantt depicts the task activity of individual processors with a horizontal bar chart in which the color of each bar indicates the current task being executed by the corresponding processor as a function of time. The Task summay displays the duration of each task as a percentage of the overall execution time and also places the duration interval of each task within the overall execution interval of the program.

• Other displays
  Critical path is similar to the space-time diagram, but uses a different color coding to highlight the longest serial thread in a parallel computation. Phase portraits illustrate the relationship over time between communication and processor use.