Article:		A New Visualization for Web Server Logs
Subject:		Better ways to see 3D
Date:		2007-02-12 17:30:13
From:		dberkholz
Try creating surfaces rather than dots. Also, if you color them by height in the Z-direction, it's a lot easier to see than with a single color. Also, you could try creating true stereo images to better see the 3D nature. Simply rotate one image by 7 degrees relative to the other. Depending on which one is on the left, you'll get either walleyed or crosseyed stereo.

Showing messages 1 through 6 of 6.

• Better ways to see 3D
  2007-02-12 17:32:21  dberkholz [View]
  
  
  Forgot to mention ... once you've made the stereo images, separate them by the distance between your eyes for optimal viewing. If they overlap, shrink them.
• Surface plot, viewing 3D
  2007-02-12 22:53:48  rajuvarghese [View]
  
  
  Thanks dberkholz for your two suggestions. I did not create a surface plot because the Z
  value is not continuous. For a small number of points it may look good. With
  larger numbers it will not display well. Color, as noted in the article, will
  be put to good use for visualizing another dimension in a forthcoming article.
  
  The jittered plot was my attempt at displaying 3D without red-blue glasses or specialized equipment.
  Since color will be used later for displaying another parameter I cannot use it for left and right images. Let me know if you had something else in mind with your suggestion.
  
  
• Surface plot, viewing 3D
  2007-02-13 18:52:24  Nick_3D [View]
  
  
  Gentlemen,
  Surface 3D visualization gives much better perception in any case as correctly mentioned by dberkholz. It is easy to understand because real objects we used to deal with in life look like surfaces rather than clouds. It is also as easy as the fact that line 2D plot look more understandable as compared to spread points. As many as several thousands points can be nicely visualized as 3D surface. Colors help a lot too as correctly mentioned by dberkholz. For more that 1 MLN points the best is still surface 3D visualization but enhanced with real-time zooming (all directions) and flyby, that is normally no problem with reasonable dedicated multidimensional graphics software package.
  For better 3D screen shots you can see this for example:
  http://www.sciencegl.com/Stock_market/Stock_market.htm
  
  What matters the most I think is a interactive way to read back the data of the interest in any point of the surfaces and at any time of navigation. Read out such as point position following mouse (XYZ), cross-section (cut), difference between markers, etc.
  Something like this
  http://www.sciencegl.com/Help_stock/Stock_Market_3D.html
  
  Thank you for your attention.
  
  
• Surface plot, viewing 3D
  2007-02-15 08:20:29  rajuvarghese [View]
  
  

  Surfaces do look better than clouds but my hesitation is due to two things:

  
  
  

  
  
  1. joining up points to form a line (2D case) and making a surface (3D case) is ok for continuous functions/data. In this case, the adjacent IP addresses may have little to do with each other and one cannot expect a smooth transition between any two points. This can apply to one of more of the axes (IP address and content). Interpolating a surface may, therefore, be fake.
  
  
  2. I have another yet dimension that I would like to bring into the picture and color is the only thing left at my disposal
  
  

  
  
  

  But you do bring up an important point: interactivity. Plots such as the ones in the article are dead objects that cannot tell you any more about the displayed data. A good interactive tool would let the user mouse over the points and describe each data point. Excel charts can do this as, I am sure, many others too. Excel has many other well-known limitations which precluded its use for decent-sized log files. I would love to try out my data with a proper 3D tool such as the one you linked to.

  
  
  

  If you have such a tool and could load it up with similar data, I would be grateful if you put up a screenshot to show that it does indeed look good.

  
  
• Surface plot, viewing 3D
  2007-02-15 18:04:44  Nick_3D [View]
  
  
  Yes, your hesitation is correct in some respect. True spread data has little or no sense in surface representation. The things are different if you have at least 2 continuous directions such as Z (volume or hits in your case) and X (time in your case) then surface works perfectly fine as it is sown in links above. 3D surf is constructed as bands or 3D tapes, interpolation only in continuous directions. Feel free to drop a line to my friends at www.ScienceGL.com to get demo, evaluation whatever. Contact Alex.
  Network graphics (http://www.sciencegl.com/)
  
• Surface plot, viewing 3D
  2007-02-13 18:48:31  Nick_3D [View]
  
  
  Gentlemen,
  Surface 3D visualization gives much better perception in any case as correctly mentioned by dberkholz. It is easy to understand because real objects we uset to seal with in life looks like surfaces rather than clouds. As many as several thousands points look still good in as 3D surface. Colors help a lot too as correctly mentioned by dberkholz. For more that 1 MLN points the best is still surface 3D visualization but enchanced with real-time zooming (all directions) and flyby, that is normally no problem with reasonable dedicated multidimentioanal graphics software package.
  For better 3D screen shots you can see this for example:
  http://www.sciencegl.com/Stock_market/Stock_market.htm
  
  What matters the most I think is a interactive way to read back the data of the interest in any point of the surfaces and at any time of navigation. Read out such as point position folowing mouse (XYZ), crossection, difference between markers, etc.
  Something like this
  http://www.sciencegl.com/Help_stock/Stock_Market_3D.html
  
  Thank you for your attention.