On Thursday 31 July 2003 04:32 am, Gordan wrote: > On Wednesday 30 July 2003 23:06, Dave Hooper wrote: > > 1. Use regular .zip because Java can do this natively > > That is so far the most compelling argument for using it. I am not sure, > however, if this convenience would outweigh the benefits of the improved > compression ration.
Ok, I was tired of people guessing about how little or much this would improve things. So I did a test. I downloaded about 2.5 megs of HTML form the web and off Freenet. I fetched large numbers of pages from each sight, so it would compress well. The statics are as follows: (all units are in bytes) The entire directory: 2617995 The tar of the directory: 2703360 The zip of the directory: 583382 Ratio: 4.48 to 1 Gzip of the tar (default): 497157 Ratio: 5.26 to 1 (compaired to the dir size) Gzip of the tar (--best): 492510 Ratio: 5.31 to 1 (compaired to the dir size) Bzip2 of the tar: 437351 Ratio: 5.98 to 1 (compaired to the dir size) This means, that if we assume this compression ratio, on a hypothetical size index, Bzip will result in enough improvement to move to the next power of 2 size 3/8ths of the time. Bottom line: On Freesites that are using HTML containers that have between 4KB and 4MB of uncompressed content, Bzips will only use 80% of bandwidth than zips. Please bare in mind that Zips are already reducing this bandwidth to about 23% of what it once was, and that space wise, this is a small part of Freenet's content. Is is taking this number down to 19% worth all the extra effort it would take? When you consider just how much better things are with compression, I think the most important thing that can be done is to insure that everything that is not already compressed, that is inserted, gets compressed. Beyond that you can easily play games in the insertion utilities, like having 2 or 3 zips for a sight and shuffle files around so they get padded less. However I think that it ultimately comes down to the fact that zips are so much essayer to implement. _______________________________________________ devl mailing list [EMAIL PROTECTED] http://hawk.freenetproject.org:8080/cgi-bin/mailman/listinfo/devl
