 |
Figure 7 plots the latencies for transmitting a
set of synthetic emails consisting of a few text paragraphs and a
variable number of image attachments each of size 100 KB over a 56 Kbps
link. The plots show results for a run that uses Outlook without any
adaptation support (Native), and two CoFi runs, one that sends
the full images (Full), and a second that uses versioning to
propagate partial-fidelity versions of the images (Partial). In
this experiment, a partial-fidelity version correspond to the initial
of the content of an image encoded in a progressive JPEG
representation.
For the Native run, we measure only the time it takes to transmit the emails between the mobile client running Outlook and an SMTP server on the other end of the bandwidth-limited link. This accounts for the time that the mobile client has to wait before disconnection in order to propagate the email. We do not include the time it takes for the SMTP server to deliver the email to the recipients, as these operations can be done asynchronously and do not require the mobile client to remain connected. Similarly, for CoFi runs we measure only the time it takes to transmit the emails between the CoFi local and remote proxies and do not include the time needed to compose and send emails to third-party email recipients, or the time it takes for CoFi-enabled recipients to read the email adaptively.
Full demonstrates that the CoFi overhead is small, averaging less than 5% over all emails. In contrast, Partial shows that progressive propagation of the attachments reduces the latency by roughly 80%. The 5% overhead in CoFi corresponds to the cost involved in parsing the email content to find its structure, exchanging the control information, and transcoding the images.