Check out the new USENIX Web site. next up previous
Next: Fourth experiment Up: Experimental results Previous: Second experiment

Third experiment

 

  figure193


Figure 8: Ideal file arrival times vs. estimated file arrival times for jsolitaire (experiment 3).

 

  figure201


Figure 9: Ideal file arrival times vs. estimated file arrival times for iconpainter (experiment 3).

In the third experiment, we measured the effectiveness of the bundlings of the rt.jar subset produced in experiment 1 on two applications not represented in the input profiles used to generate those bundlings. The applications were jsolitaire (a Solitaire card game applet) and iconpainter (an icon editor). Note that the iconpainter used four class files not loaded in any of the original input profiles; we added these to the bundlings as bundles of size 1 (i.e., containing a single class file).

Figures 8 and 9 show the expected vs. ideal file arrival times for jsolitaire and iconpainter, respectively. The arrival times for the bundlings are somewhat further from ideal than in experiment 1, where the profiles tested were members of the set of profiles used to generate the bundlings. This shows that bundling works best for applications which are represented in the input profiles used to generate the bundlings. However, the time for all of the files to arrive for the loose bundlings is only about 40% later than for the ideal case. Considering that bundling typically achieves twice the compression of Jar archives, this overhead may be acceptable in some situations.



David Hovemeyer
Tue Feb 27 18:43:09 EST 2001