CoFi-enabled PowerPoint propagates modifications progressively by saving back just subsets of the modified slides or embedded objects, or by transcoding embedded images into a progressive JPEG representation and saving just portions of the images' data. We implemented an adaptation policy that propagates modifications every time the user saves the document. The policy propagates the text content of any new or modified slides and transcoded versions of new or modified embedded images. Image fidelity refinements are then propagated on every subsequent save request until all images at the CoFi remote proxy reach full fidelity.
 |
We evaluate the effectiveness of progressive update propagation by measuring the latency for saving modifications to a set of synthetic PowerPoint documents. We constructed our synthetic documents by replicating a single slide that contained 4 KB of text and a 80 KB image.
Figure 9 shows latency measurements for saving PowerPoint documents with up to 50 slides over a 56 Kb/sec network link. The figure shows latency results for transferring the documents over FTP and adaptation policies that use subsetting and versioning to reduce the data traffic. The FTP measurements give us a baseline for the time it takes to transfer the full document without any adaptation. We use this baseline to determine the effectiveness of our adaptation policies.
Figure 9 plots the results of 5 experiments that
use subsetting to reduce latency. The numbers on the right hand side
of the plot, next to each line, show the proportion of document slides
that was saved back to the remote proxy. In this manner, the top most
line corresponds to documents that were saved in their entirety, while
the lowest subsetting line corresponds to documents where only
modifications to 20% of the slides were saved. In all experiments,
we assume that both the slide's text and single image were modified
and had to be saved back. The top most subsetting line, which
corresponds to saving the full document, shows that the CoFi overhead
is small, averaging less than 5% over all documents. In contrast, all
other subsetting experiments show significant reductions in upload
latency. The last five lines in Figure 9 show the
results for an adaptation policy that uses subsetting and versioning
of images to further reduce upload latency. This policy converts
images embedded in slides into a progressive JPEG representation and
transfers only the initial 
of the image's data; achieving even
larger reductions in upload latency.