SimGrid  3.16
Versatile Simulation of Distributed Systems
Visualization and Statistical Analysis

SimGrid comes with an extensive support to trace and register what happens during the simulation, so that it can be either visualized or statistically analysed after the simulation.

This tracing is widely used to observe and understand the behavior of parallel applications and distributed algorithms. Usually, this is done in a two-step fashion: the user instruments the application and the traces are analyzed after the end of the execution. The analysis can highlights unexpected behaviors, bottlenecks and sometimes can be used to correct distributed algorithms. The SimGrid team has instrumented the library in order to let users trace their simulations and analyze them. This part of the user manual explains how the tracing-related features can be enabled and used during the development of simulators using the SimGrid library.

Tracing categories functions

The SimGrid library is instrumented so users can trace the platform utilization using MSG, SimDAG and SMPI interfaces. It registers how much power is used for each host and how much bandwidth is used for each link of the platform. The idea with this type of tracing is to observe the overall view of resources utilization in the first place, especially the identification of bottlenecks, load-balancing among hosts, and so on.

Another possibility is to trace resource utilization by categories. Categorized resource utilization tracing gives SimGrid users to possibility to classify MSG and SimDAG tasks by category, tracing resource utilization for each of the categories. The functions below let the user declare a category and apply it to tasks. The tasks that are not classified according to a category are not traced. Even if the user does not specify any category, the simulations can still be traced in terms of resource utilization by using a special parameter that is detailed below (see section Tracing configuration Options).

Tracing marks functions

Tracing user variables functions

For hosts:

For links:

For links, but use source and destination to get route:

Tracing configuration Options

To check which tracing options are available for your simulator, you can just run it with the option

--help-tracing 

to get a very detailed and updated explanation of each tracing parameter. These are some of the options accepted by the tracing system of SimGrid, you can use them by running your simulator with the –cfg= switch:

  • tracing : Safe switch. It activates (or deactivates) the tracing system. No other tracing options take effect if this one is not activated.
    --cfg=tracing:yes
    
  • tracing/categorized : It activates the categorized resource utilization tracing. It should be enabled if tracing categories are used by this simulator.
    --cfg=tracing/categorized:yes
    
  • tracing/uncategorized : It activates the uncategorized resource utilization tracing. Use it if this simulator do not use tracing categories and resource use have to be traced.
    --cfg=tracing/uncategorized:yes
    
  • tracing/filename : A file with this name will be created to register the simulation. The file is in the Paje format and can be analyzed using Viva or Paje visualization tools. More information can be found in these webpages: http://github.com/schnorr/viva/ http://github.com/schnorr/pajeng/
    --cfg=tracing/filename:mytracefile.trace
    
    If you do not provide this parameter, the trace file will be named simgrid.trace.
  • tracing/smpi : This option only has effect if this simulator is SMPI-based. Traces the MPI interface and generates a trace that can be analyzed using Gantt-like visualizations. Every MPI function (implemented by SMPI) is transformed in a state, and point-to-point communications can be analyzed with arrows.
    --cfg=tracing/smpi:yes
    
  • tracing/smpi/group : This option only has effect if this simulator is SMPI-based. The processes are grouped by the hosts where they were executed.
    --cfg=tracing/smpi/group:yes
    
  • tracing/smpi/computing : This option only has effect if this simulator is SMPI-based. The parts external to SMPI are also outputted to the trace. Provides better way to analyze the data automatically.
    --cfg=tracing/smpi/computing:yes
    
  • tracing/smpi/internals : This option only has effect if this simulator is SMPI-based. Display internal communications happening during a collective MPI call.
    --cfg=tracing/smpi/internals:yes
    
  • tracing/smpi/display-sizes : This option only has effect if this simulator is SMPI-based. Display the sizes of the messages exchanged in the trace, both in the links and on the states. For collective, size means the global size of data sent by the process in general.
    --cfg=tracing/smpi/display-sizes:yes
    
  • tracing/smpi/sleeping : TODO
    TODO
    
  • tracing/smpi/format : TODO
    TODO
    
  • tracing/smpi/format/ti-one-file : TODO
    TODO
    
  • tracing/msg/vm : TODO
    TODO
    
  • tracing/msg/process : This option only has effect if this simulator is MSG-based. It traces the behavior of all categorized MSG processes, grouping them by hosts. This option can be used to track process location if this simulator has process migration.
    --cfg=tracing/msg/process:yes
    
  • tracing/buffer : This option put some events in a time-ordered buffer using the insertion sort algorithm. The process of acquiring and releasing locks to access this buffer and the cost of the sorting algorithm make this process slow. The simulator performance can be severely impacted if this option is activated, but you are sure to get a trace file with events sorted.
    --cfg=tracing/buffer:yes
    
  • tracing/onelink-only : This option changes the way SimGrid register its platform on the trace file. Normally, the tracing considers all routes (no matter their size) on the platform file to re-create the resource topology. If this option is activated, only the routes with one link are used to register the topology within a netzone. Routes among netzones continue to be traced as usual.
    --cfg=tracing/onelink-only:yes
    
  • tracing/disable-link : TODO
    TODO
    
  • tracing/disable-power : TODO
    TODO
    
  • tracing/disable-destroy : Disable the destruction of containers at the end of simulation. This can be used with simulators that have a different notion of time (different from the simulated time).
    --cfg=tracing/disable-destroy:yes
    
  • tracing/basic : Some visualization tools are not able to parse correctly the Paje file format. Use this option if you are using one of these tools to visualize the simulation trace. Keep in mind that the trace might be incomplete, without all the information that would be registered otherwise.
    --cfg=tracing/basic:yes
    
  • tracing/comment : Use this to add a comment line to the top of the trace file.
    --cfg=tracing/comment:my_string
    
  • tracing/comment-file : Use this to add the contents of a file to the top of the trace file as comment.
    --cfg=tracing/comment-file:textual_file.txt
    
  • tracing/precision : This option determines the precision of timings stored in the trace file. Make sure you set Numerical precision of the platform models to at least the same value as this option! (Traces cannot be more accurate than the simulation; they can be less accurate, though.)

The following example will give you a precision of E-10 in the trace file:

--cfg=tracing/precision:10
  • tracing/platform : TODO
    TODO
    
  • tracing/platform/topology : TODO
    TODO
    
  • viva/categorized : This option generates a graph configuration file for Viva considering categorized resource utilization.
    --cfg=viva/categorized:graph_categorized.plist
    
  • viva/uncategorized : This option generates a graph configuration file for Viva considering uncategorized resource utilization.
    --cfg=viva/uncategorized:graph_uncategorized.plist
    

Please pass

--help-tracing 

to your simulator for the updated list of tracing options.

Case studies

Some scenarios that might help you decide which tracing options you should use to analyze your simulator.

  • I want to trace the resource utilization of all hosts and links of the platform, and my simulator does not use the tracing API. For that, you can run a uncategorized trace with the following parameters (it will work with any Simgrid simulator):
    ./your_simulator \
              --cfg=tracing:yes \
              --cfg=tracing/uncategorized:yes \
              --cfg=tracing/filename:mytracefile.trace \
              --cfg=viva/uncategorized:uncat.plist
    
  • I want to trace only a subset of my MSG (or SimDAG) tasks. For that, you will need to create tracing categories using the TRACE_category (...) function (as explained above), and then classify your tasks to a previously declared category using the MSG_task_set_category (...) (or SD_task_set_category (...) for SimDAG tasks). After recompiling, run your simulator with the following parameters:
    ./your_simulator \
              --cfg=tracing:yes \
              --cfg=tracing/categorized:yes \
              --cfg=tracing/filename:mytracefile.trace \
              --cfg=viva/categorized:cat.plist
    

Example of Instrumentation

A simplified example using the tracing mandatory functions.

int main (int argc, char **argv)
{
  MSG_init (&argc, &argv);

  //(... after deployment ...)

  //note that category declaration must be called after MSG_create_environment
  TRACE_category_with_color ("request", "1 0 0");
  TRACE_category_with_color ("computation", "0.3 1 0.4");
  TRACE_category ("finalize");

  msg_task_t req1 = MSG_task_create("1st_request_task", 10, 10, NULL);
  msg_task_t req2 = MSG_task_create("2nd_request_task", 10, 10, NULL);
  msg_task_t req3 = MSG_task_create("3rd_request_task", 10, 10, NULL);
  msg_task_t req4 = MSG_task_create("4th_request_task", 10, 10, NULL);
  MSG_task_set_category (req1, "request");
  MSG_task_set_category (req2, "request");
  MSG_task_set_category (req3, "request");
  MSG_task_set_category (req4, "request");

  msg_task_t comp = MSG_task_create ("comp_task", 100, 100, NULL);
  MSG_task_set_category (comp, "computation");

  msg_task_t finalize = MSG_task_create ("finalize", 0, 0, NULL);
  MSG_task_set_category (finalize, "finalize");

  //(...)

  MSG_clean();
  return 0;
}

Analyzing SimGrid Simulation Traces

A SimGrid-based simulator, when executed with the correct parameters (see above) creates a trace file in the Paje file format holding the simulated behavior of the application or the platform. You have several options to analyze this trace file:

  • Dump its contents to a CSV-like format using pj_dump (see PajeNG's wiki on pj_dump and more generally the PajeNG suite) and use gnuplot to plot resource usage, time spent on blocking/executing functions, and so on. Filtering capabilities are at your hand by doing grep, with the best regular expression you can provide, to get only parts of the trace (for instance, only a subset of resources or processes).
  • Derive statistics from trace metrics (the ones built-in with any SimGrid simulation, but also those metrics you injected in the trace using the TRACE module) using the R project and all its modules. You can also combine R with ggplot2 to get a number of high quality plots from your simulation metrics. You need to pj_dump the contents of the SimGrid trace file to use R.
  • Visualize the behavior of your simulation using classic space/time views (gantt-charts) provided by the PajeNG suite and any other tool that supports the Paje file format. Consider this option if you need to understand the causality of your distributed simulation.
  • Visualize the behavior of your simulation with treemaps (specially if your simulation has a platform with several thousand resources), provided by the Viva visualization tool. See Viva's wiki for further details on what is a treemap and how to use it.
  • Correlate the behavior of your simulator with the platform topology with an interactive, force-directed, and hierarchical graph visualization, provided by Viva. Check Viva's wiki for further details. This research report, published at ISPASS 2013, has a detailed description of this visualization technique.
  • You can also check our online tutorial section that contains a dedicated tutorial with several suggestions on how to use the tracing infrastructure. Look for the SimGrid User::Visualization 101 tutorial.
  • Ask for help on the simgrid-user@lists.gforge.inria.fr mailing list, giving us a detailed explanation on what your simulator does and what kind of information you want to trace. You can also check the mailing list archive for old messages regarding tracing and analysis.

Viva Visualization Tool

This subsection describe some of the concepts regarding the Viva Visualization Tool and its relation with SimGrid traces. You should refer to Viva's website for further details on all its visualization techniques.

Time Slice

The analysis of a trace file using the tool always takes into account the concept of the time-slice. This concept means that what is being visualized in the screen is always calculated considering a specific time frame, with its beggining and end timestamp. The time-slice is configured by the user and can be changed dynamically through the window called Time Interval that is opened whenever a trace file is being analyzed. Users are capable to select the beggining and size of the time slice.

Hierarchical Graph View

As stated above (see section Analyzing SimGrid Simulation Traces), one possibility to analyze SimGrid traces is to use Viva's graph view with a graph configuration to customize the graph according to the traces. A valid graph configuration (we are using the non-XML Property List Format to describe the configuration) can be created for any SimGrid-based simulator using the –cfg=viva/uncategorized:graph_uncategorized.plist or –cfg=viva/categorized:graph_categorized.plist (if the simulator defines resource utilization categories) when executing the simulation.

Basic Graph Configuration

The basic description of the configuration is as follows:

{
  node = (LINK, HOST, );
  edge = (HOST-LINK, LINK-HOST, LINK-LINK, );

The nodes of the graph will be created based on the node parameter, which in this case is the different "HOST"s and "LINK"s of the platform used to simulate. The edge parameter indicates that the edges of the graph will be created based on the "HOST-LINK"s, "LINK-HOST"s, and "LINK-LINK"s of the platform. After the definition of these two parameters, the configuration must detail how the nodes (HOSTs and LINKs) should be drawn.

For that, the configuration must have an entry for each of the types used. For HOST, as basic configuration, we have:

  HOST = {
    type = square;
    size = power;
    values = (power_used);
  };

The parameter size indicates which variable from the trace file will be used to define the size of the node HOST in the visualization. If the simulation was executed with availability traces, the size of the nodes will be changed according to these traces. The parameter type indicates which geometrical shape will be used to represent HOST, and the values parameter indicates which values from the trace will be used to fill the shape.

For LINK we have:

  LINK = {
    type = rhombus;
    size = bandwidth;
    values = (bandwidth_used);
  };
}

The same configuration parameters are used here: type (with a rhombus), the size (whose value is from trace's bandwidth variable) and the values.

Customizing the Graph Representation

Viva is capable to handle a customized graph representation based on the variables present in the trace file. In the case of SimGrid, every time a category is created for tasks, two variables in the trace file are defined: one to indicate node utilization (how much power was used by that task category), and another to indicate link utilization (how much bandwidth was used by that category). For instance, if the user declares a category named request, there will be variables named prequest and a brequest (p for power and b for bandwidth). It is important to notice that the variable prequest in this case is only available for HOST, and brequest is only available for LINK. Example: suppose there are two categories for tasks: request and compute. To create a customized graph representation with a proportional separation of host and link utilization, use as configuration for HOST and LINK this:

  HOST = {
    type = square;
    size = power;
    values = (prequest, pcomputation);
  };
  LINK = {
    type = rhombus;
    size = bandwidth;
    values = (brequest, bcomputation);
  };

This configuration enables the analysis of resource utilization by MSG tasks through the identification of load-balancing issues and network bottlenecks, for instance.