On-the-fly compression is a variation on downloading individual files. Files are requested individually by the client, and transferred from the server over a compressed stream. Any suitable compression algorithm could be used; zlib [1] is an obvious choice since it is a standard part of the Java API (in the java.util.zip package)Note that for on-the-fly compression to work, the compression algorithm must have the ability to flush the compressed stream at arbitrary points (i.e., the file boundaries). Zlib has this ability, but it is not implemented in the java.util.zip package..
On-the-fly compression has the advantages of downloading individual files: only the files requested by the client are transferred, and the files arrive in the order needed. The fact that the entire stream of files is compressed cumulatively means that a high degree of sharing of redundant information between files is possible, resulting in compression ratios higher than those achieved by simply compressing each file individually.
However, on-the-fly compression has the disadvantage that it can require considerable CPU time on the server to perform the compression. For example, a 333 MHz Sun Ultra 5 can compress about 2.5 megabytes per second using zlib at the lowest compression level, and about 0.5 megabytes per second at the highest compression level. Considering that server will have more work to do at runtime in addition to compression, the sustainable throughput will be somewhat less. This limits the scalability of on-the-fly compression using zlib.
Another disadvantage of on-the-fly compression is that it does not use pre-sending or prefetching, meaning that the network latency cost is paid for each file requested.