Prototype "improved_chk_index"

Alexander Belchenko bialix at
Thu Oct 29 06:23:29 GMT 2009

John Arbash Meinel пишет:
>    The bad of it, is that the file is complete garbage in a text editor.
>    Though I have to say, btree indexes are zlib compressed on 4k
>    boundaries, so their content is just as opaque. (and you can't just
>    'gunzip' the index because it is zlib not gz, and based on repeated
>    compression, rather than compressing the whole stream.)

IMO if you trying to get the maximum performance and minimum memory size requirements you have to
throw away any text-based formats and use only fine tuned binary data formats. Of course you need to
spend some time to write specific debugging tools and data viewers. But it will gain you a lot anyway.

In my embedded system practice when every byte and every clock cycle counts binary protocols beat
text-based every time.

More information about the bazaar mailing list