RoomEditorApp Idling and Benchmarking Timer

I solved my two RoomEditorApp Idling issues!

They were especially worrying due to the fact that I am planning to present this application in my Autodesk university class DV1736 on Cloud-Based, Real-Time, Round-Trip, 2D Revit Model Editing on Any Mobile Device.

By the way, you should now enrol in the AU classes you are interested in – unless you already did :-)

Go to the AU class catalogue, sign in, click 'View All Classes', use the filters or keyword search to locate an event you want to schedule and add it to your personal schedule using the scheduling options and blue plus sign.

Here are my four classes, updated from the AU class catalogue preview version:

Back to my RoomEditorApp, though, and the nitty-gritty details of resolving my Idling issues.

RoomEditorApp Idling Issues

I recently wrote about my updated Revit 2014 version of the RoomEditorApp, the Revit add-in part of my cloud-based real-time round-trip 2D Revit model editing application, the RoomEditorApp GitHub repository and its architecture and external application implementation.

In closing, I mentioned two worrying Idling related problems:

Happily, and somewhat surprisingly, I yesterday resolved them both.

The issue with the unsubscription was easy, as I recently discovered and discussed the cause.

I was still worried that I might have to switch back to Revit 2013 for the demo due to the unresponsiveness, however.

Luckily, I thought of adding some benchmarking instrumentation code to find out where the issue was, and that helped find and fix the root cause.

It was my own fault :-)

Here are the steps:

Fixing the RoomEditorApp Idling Unsubscription

My attempts to unsubscribe from the Idling event were failing due to the fact that each new external command invocation generates a new different class instance. The unsubscribing instance was therefore providing a different member method to unsubscribe than the original one used to subscribe, so the original just kept going, as explained in the discussion on singleton application versus multiple command instances.

That was really easy to fix, since I only had to modify the external application to keep track of the handler used to subscribe, and reuse that specific stored method delegate when unsubscribing again.

I published the fix in release 2014.0.0.16.

Adding a Benchmarking Timer

While I was at it, I decided to spend at least a little time trying to understand the cause for this unresponsiveness.

But where to start?

Well, first of all, I needed to understand where the time was actually being consumed.

To do so, I looked at my timer code for benchmarking and the JtTimer class that I used to create some filtered element collector benchmarks.

Here is the spruced up version that I ended up using for the following tests:

/// <summary>
/// Performance timer for profiling purposes.
/// For a full description, please refer to
/// http://thebuildingcoder.typepad.com/blog/2010/03/performance-profiling.html
/// </summary>
public class JtTimer : IDisposable
{
  #region Internal TimeRegistry class
  class TimeRegistry
  {
    #region Internal data and helper methods
    class Entry
    {
      public double Time { get; set; }
      public int Calls { get; set; }
    }
 
    static Dictionary<string, Entry> _collection
      = new Dictionary<string, Entry>();
 
    /// <summary>
    /// Return the percentage based on total time.
    /// </summary>
    /// <param name="value">value</param>
    /// <param name="totalTime">total time</param>
    /// <returns></returns>
    static double GetPercent(
      double value,
      double totalTime )
    {
      return 0 == totalTime
        ? 0
        : Math.Round( value * 100 / totalTime, 2 );
    }
    #endregion // Internal data and helper methods
 
    /// <summary>
    /// Add new duration for specified key.
    /// </summary>
    public static void AddTime(
      string key,
      double duration )
    {
      Entry e;
      if( _collection.ContainsKey( key ) )
      {
        e = _collection[key];
      }
      else
      {
        e = new Entry();
        _collection.Add( key, e );
      }
      e.Time += duration;
      ++e.Calls;
    }
 
    /// <summary>
    /// Write the report of the results to a text file.
    /// </summary>
    public static void WriteResults(
      string description,
      double totalTime )
    {
      // Set up text file path:
 
      string strReportPath = Path.Combine(
        Path.GetTempPath(), "PerformanceReport.txt" );
 
      FileStream fs = new FileStream( strReportPath,
        FileMode.OpenOrCreate, FileAccess.Write );
 
      StreamWriter streamWriter = new StreamWriter( fs );
      streamWriter.BaseStream.Seek( 0, SeekOrigin.End );
 
      // Sort output by percentage of total time used:
 
      List<string> lines = new List<string>(
        _collection.Count );
 
      foreach( KeyValuePair<string, Entry> pair
        in _collection )
      {
        Entry e = pair.Value;
 
        lines.Add( string.Format(
          "{0,10:0.00}%{1,10:0.00}{2,8}   {3}",
          GetPercent( e.Time, totalTime ),
          Math.Round( e.Time, 2 ),
          e.Calls,
          pair.Key ) );
      }
      lines.Sort();
 
      string header
        = " Percentage   Seconds   Calls   Process";
 
      int n = Math.Max( header.Length,
        lines.Max<string>( x => x.Length ) );
 
      if( null != description
        && 0 < description.Length )
      {
        n = Math.Max( n, description.Length );
        header = description + "\r\n" + header;
      }
      string separator = "-";
      while( 0 < n-- )
      {
        separator += "-";
      }
      streamWriter.WriteLine( separator );
      streamWriter.WriteLine( header );
      streamWriter.WriteLine( separator );
 
      foreach( string line in lines )
      {
        streamWriter.WriteLine( line );
      }
      streamWriter.WriteLine( separator + "\r\n" );
      streamWriter.Close();
      fs.Close();
      Process.Start( strReportPath );
      _collection.Clear();
    }
  }
  #endregion // Internal TimeRegistry class
 
  string _key;
  Stopwatch _timer;
  double _duration = 0;
 
  /// <summary>
  /// Performance timer constructor.
  /// </summary>
  /// <param name="what_are_we_testing_here">
  /// Key describing code to be timed</param>
  public JtTimer( string what_are_we_testing_here )
  {
    Restart( what_are_we_testing_here );
  }
 
  /// <summary>
  /// Automatic disposal when the the using statement
  /// block finishes: the timer is stopped and the
  /// time is registered.
  /// </summary>
  void IDisposable.Dispose()
  {
    Stop();
  }
 
  /// <summary>
  /// Write and display a report of the timing
  /// results in a text file.
  /// </summary>
  public void Report( string description )
  {
    TimeRegistry.WriteResults(
      description, _duration );
  }
 
  /// <summary>
  /// Restart the measurement from scratch.
  /// </summary>
  public void Restart( string what_are_we_testing_here )
  {
    _key = what_are_we_testing_here;
    _timer = Stopwatch.StartNew();
  }
 
  /// <summary>
  /// Stop the timer.
  /// </summary>
  public void Stop()
  {
    _timer.Stop();
    _duration = _timer.Elapsed.TotalSeconds;
    TimeRegistry.AddTime( _key, _duration );
  }
}

Benchmarking Instrumentation

To use the timer, I equipped various methods of interest with a using call to its constructor, e.g. like this:

  /// <summary>
  /// Return the last sequence number.
  /// </summary>
  public int LastSequenceNumber
  {
    get
    {
      using( JtTimer pt = new JtTimer(
        "LastSequenceNumber" ) )
      {
        ChangeOptions opt = new ChangeOptions();
 
        CouchChanges<DbFurniture> changes
          = _db.GetChanges<DbFurniture>( opt );
 
        CouchChangeResult<DbFurniture> r
          = changes.Results.Last<
            CouchChangeResult<DbFurniture>>();
 
        return r.Sequence;
      }
    }
  }

As you can see, I just add the using statement and a call to the JtTimer constructor to each method I would like to track performance for, nothing else.

All the rest is handled automatically by the JtTimer destructor and other internal methods.

I also added a call to instantiate a top-level timer keeping track of the total time and reporting results each time subscription is started and stopped, respectively, like this:

class App : IExternalApplication
{
  /// <summary>
  /// Subscription debugging timer.
  /// </summary>
  static JtTimer _timer = null;
 
  /// <summary>
  /// Toggle on and off subscription to 
  /// automatic cloud updates.
  /// </summary>
  public static void ToggleSubscription(
    EventHandler<IdlingEventArgs> handler )
  {
    if( Subscribed )
    {
      Debug.Print( "Unsubscribing..." );
      _uiapp.Idling -= _handler;
      _handler = null;
      _buttons[3].ItemText = _subscribe;
      _timer.Stop();
      _timer.Report( "Subscription timing" );
      _timer = null;
      Debug.Print( "Unsubscribed." );
    }
    else
    {
      Debug.Print( "Subscribing..." );
      _uiapp.Idling += handler;
      _handler = handler;
      _buttons[3].ItemText = _unsubscribe;
      _timer = new JtTimer( "Subscription" );
      Debug.Print( "Subscribed." );
    }
  }
 

Here are all the occurrences of the string "JtTimer" in the entire project (copy and paste somewhere or view source to see truncated lines in full):

Find all "JtTimer", Find Results 1, "Entire Solution", "*.cs"
  App.cs(34):           static JtTimer _timer = null;
  App.cs(234):          _timer = new JtTimer( "Subscription" );
  CmdSubscribe.cs(43):  using( JtTimer pt = new JtTimer( "OnIdling" ) )
  DbUpdater.cs(59):     using( JtTimer pt = new JtTimer( "DbUpdater ctor" ) )
  DbUpdater.cs(166):    using( JtTimer pt = new JtTimer( "UpdateBim" ) )
  JtTimer.cs(3):        // JtTimer.cs - performance profiling timer
  JtTimer.cs(25):       public class JtTimer : IDisposable
  JtTimer.cs(157):      public JtTimer( string what_are_we_testing_here )
  Properties\AssemblyInfo.cs(61):  // 2013-11-18 - 2014.0.0.17 - added JtTimer
  RoomEditorDb.cs(24):  using( JtTimer pt = new JtTimer( "RoomEditorDb ctor" ) )
  RoomEditorDb.cs(52):  using( JtTimer pt = new JtTimer( "LastSequenceNumber" ) )
  RoomEditorDb.cs(81):  using( JtTimer pt = new JtTimer( "LastSequenceNumberChanged" ) )
  Matching lines: 12
  Matching files: 6
  Total files searched: 21

Initial Benchmarking Results and Conclusion

In the beginning, I was not stopping the top-level timer properly, so I had no grand total, and therefore no percentages.

The very first result looked like this, with just two timers active, showing that I was headed in the right direction:

------------------------------------------
Subscription timing
 Percentage   Seconds   Calls   Process
------------------------------------------
      0.00%      0.04       1   UpdateBim
      0.00%     41.73     420   OnIdling
------------------------------------------

This already tells me that the Idling event handler is consuming a lot of time, whereas the actual BIM update is quick.

Adding a few more timers here and there starts producing useful results:

---------------------------------------------------
Subscription timing
 Percentage   Seconds   Calls   Process
---------------------------------------------------
      0.00%      0.00       1   DbUpdater ctor
      0.00%      0.05       1   UpdateBim
      0.00%      3.27      51   RoomEditorDb ctor
      0.00%     76.40     480   OnIdling
      0.00%     80.35      49   LastSequenceNumber
---------------------------------------------------

The LastSequenceNumber property is eating up absolutely all the time!

It is called once before we start subscribing to the Idling event, and then from inside the event handler.

All of the Idling event handler time is consumed by calls to this property.

I already showed the method implementation as an example of the benchmarking instrumentation.

Looking more closely at this method, I noted that it retrieves all the furniture documents from the database in order to determine the most recent sequence number. The sequence number is used to afterwards retrieve all database changes that occurred after a certain point.

This call will obviously take longer and longer time the more entries we add to the database.

Improving the CouchDB Sequence Number Query

In the calls made to the LastSequenceNumber property from the Idling event handler, it is used only to check whether any new changes occurred.

At that point, we already have an initial sequence number to start from, and can ask the database to return only changes that occurred after that point in time, using a 'since' argument, e.g. like this:

  /// <summary>
  /// Determine whether the given sequence number
  /// matches the most up-to-date status.
  /// </summary>
  public bool LastSequenceNumberChanged( int since )
  {
    using( JtTimer pt = new JtTimer(
      "LastSequenceNumberChanged" ) )
    {
      ChangeOptions opt = new ChangeOptions();
 
      opt.Since = since;
      opt.IncludeDocs = false;
 
      CouchChanges<DbFurniture> changes
        = _db.GetChanges<DbFurniture>( opt );
 
      CouchChangeResult<DbFurniture> r
        = changes.Results.LastOrDefault<
          CouchChangeResult<DbFurniture>>();
 
      Debug.Assert( null == r || since < r.Sequence,
        "expected monotone growing sequence number" );
 
      return null != r && since < r.Sequence;
    }
  }

Replacing the call to the LastSequenceNumber property by a call to the new LastSequenceNumberChanged method in the Idling event handler shows a huge improvement:

----------------------------------------------------------
Subscription timing
 Percentage   Seconds   Calls   Process
----------------------------------------------------------
      0.00%      0.00       1   DbUpdater ctor
      0.00%      0.11       1   UpdateBim
      0.00%      3.25    1741   RoomEditorDb ctor
      0.00%      5.35       1   LastSequenceNumber
      0.00%     19.79    1738   LastSequenceNumberChanged
      0.00%     23.55    1951   OnIdling
----------------------------------------------------------

Now a much larger number of calls to the RoomEditorDb constructor and the LastSequenceNumberChanged method can be completed in much less time.

The one and only initial call to LastSequenceNumber is still awfully expensive and could probably also be improved, but the LastSequenceNumberChanged method is quick, and the whole Idling handling is immediately much more responsive.

Twiddling with Idling Responsiveness

I added the Stop method to the timer class and call when unsubscribing from Idling, as shown above in the benchmarking instrumentation code.

That switches on the top-level timer providing the 100% measurement and enables the percentage calculations, like this:

----------------------------------------------------------
Subscription timing
 Percentage   Seconds   Calls   Process
----------------------------------------------------------
      0.00%      0.00       1   DbUpdater ctor
      0.25%      0.05       1   UpdateBim
      8.85%      1.67     486   LastSequenceNumberChanged
     16.06%      3.03     489   RoomEditorDb ctor
     17.05%      3.22   48667   OnIdling
     27.89%      5.27       1   LastSequenceNumber
    100.00%     18.89       1   Subscription
----------------------------------------------------------

With a more responsive behaviour, I am able to experiment a bit. For instance, I removed the stopwatch in the Idling event handler again, call SetRaiseWithoutDelay immediately, and ignore 99 out of hundred calls to the Idling event handler. As you can see, the Idling handler is called a hundred times for each check of the LastSequenceNumberChanged property:

----------------------------------------------------------
Subscription timing
 Percentage   Seconds   Calls   Process
----------------------------------------------------------
      0.00%      0.00       2   DbUpdater ctor
      0.11%      0.08       2   UpdateBim
      4.46%      3.23    1693   RoomEditorDb ctor
      8.35%      6.05       1   LastSequenceNumber
     21.72%     15.73    1689   LastSequenceNumberChanged
     27.39%     19.84  168900   OnIdling
    100.00%     72.44       1   Subscription
----------------------------------------------------------

The one single call to the LastSequenceNumber property costs more time than almost 500 calls to LastSequenceNumberChanged and almost 50000 calls to the Idling event handler.

Whew.

Download

This application lives in the RoomEditorApp GitHub repository and the versions discussed above are:

I hope that this is of interest to you for numerous reasons.

The most important lesson should be: please benchmark your code!