Some FSFS repo stats

[Stefan Fuhrmann <st...@wandisco.com>]

Hi there,

For those of you that are facts and numbers crazy,
I attached some data size info for 3 large FSFS
repositories. They are 1.8-format mirrors of the
Apache, KDE and wordpress repositories. I used
my new fsfs-stats tool to extract the info.

Some of my findings:

* Apache: lots of large zip files added lately
  (low overall compression rate but tool does not
   list zip files etc. as the reason - yet)
* KDE: still larger then Apache with an excellent
  compression ratio (lots of large .po files); >1TB
* Wordpress: directory compression eliminated
  directory storage overhead (5000% => <10%)

* rep sharing is most effective when you have many
  "casual" users (> factor 2 in wordpress; 25% savings
  for Apache; insignificant for KDE since po files
  are not shared / identical between branches)
* noderevs + changes list takes up 10..30% of
  the total repo size, i.e. actual content already
  well compressed

* more different file props reps than I thought
  (probably due to per-file old merge info)
* >50% of all nodes in Apache repo have props
* rep sharing + deltification brings prop info down
  to ~10 bytes / rev for Apache

-- Stefan^2.

-- 
Certified & Supported Apache Subversion Downloads:
*

http://www.wandisco.com/subversion/download
*