Although a detailed performance analysis is beyond the scope of this
paper, our modified client is well suited to handle data gathering
tasks. As was previously mentioned, we started with an already
modified version of the code that was able to perform exact traces on
each and every client access. In addition to this, our current
modifications incorporate some performance measures as well, although
not quite as detailed. We only wanted some general measures of how
the system is performing, and how well the prefetching is working.
To this end, we keep track of the following items: the total number
of bytes accessed, the number of bytes hit in the memory cache, the
number of bytes hit in the disk cache, the total number of bytes
prefetched, the number of bytes hit in the disk cache due to
prefetching, and the total number of sessions recorded. All this data
is stored in a file called .mosaic-data and is updated at the
end of each session, so the data count can span multiple sessions.
Through these data items, it is possible to determine the amount of
bandwidth being wasted due to prefetching as well as the amount of
bandwidth saved over the original Mosaic code.
Unfortunately, at this time we have not been able to do enough testing
to produce any concrete results with regards to the prefetching
performance, but a detailed analysis will be presented in future work.