rsync - Jun 2015 - rsync very slow with large include/exclude file list

I investigated the rsync code and found the reason why.
For every file in the source, it searches the entire filter-list looking to
see if that filename is on the exclude/include list.  Most aren't, so it
compares (350K - 72K) * 72K names (the non-listed files) plus (72K * 72K/2)
names (the ones that are listed), for a total of about  22,608,000,000
strcmp's.  That's 22 BILLION comparisons. (I may have left off a zero
there, it might be 220 B).

I'm working on a fix to improve this.  The first phase was to just improve
the existing code without changing the methodology.
The set I've been testing with is local-local machine, dry-run, 216K files
in the source directory, 25,000 files in the exclude-from list.
The original rsync takes 488 seconds.
The improved code takes 300 seconds.

The next phase was to improve the algorithm of handling large
filter_lists.  Change the unsorted linear search to a sorted binary search
(skiplist).
This improved code takes 2 seconds.

The original code does 4,492,304,682 strcmp's.
The fully improved code does 6,472,564.  98.5% fewer.

I am cleaning up the code and will submit a patchfile soon.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.samba.org/pipermail/rsync/attachments/20150615/f0692795/attachment.html>

This is similar to using fuzzy / -y in a large directory. O(n^2) behaviour
occurs and can be incredibly slow. No caching of md5's for the directory
occurs, it would seem (or even so, there are O(N^2) comparisons).

/kc


On Mon, Jun 15, 2015 at 06:02:14PM -0500, ray vantassle said:
  >   I investigated the rsync code and found the reason why.
  >   For every file in the source, it searches the entire filter-list
looking
  >   to see if that filename is on the exclude/include list.** Most
aren't, so
  >   it compares (350K - 72K) * 72K names (the non-listed files) plus (72K *
  >   72K/2) names (the ones that are listed), for a total of about**
  >   22,608,000,000 strcmp's.** That's 22 BILLION comparisons. (I
may have left
  >   off a zero there, it might be 220 B).
  >
  >   I'm working on a fix to improve this.** The first phase was to just
  >   improve the existing code without changing the methodology.
  >   The set I've been testing with is local-local machine, dry-run,
216K files
  >   in the source directory, 25,000 files in the exclude-from list.
  >   The original rsync takes 488 seconds.
  >   The improved code takes 300 seconds.
  >
  >   The next phase was to improve the algorithm of handling large
  >   filter_lists.** Change the unsorted linear search to a sorted binary
  >   search (skiplist).
  >   This improved code takes 2 seconds.
  >
  >   The original code does 4,492,304,682 strcmp's.
  >   The fully improved code does 6,472,564.** 98.5% fewer.
  >
  >   I am cleaning up the code and will submit a patchfile soon.

  >-- 
  >Please use reply-all for most replies to avoid omitting the mailing list.
  >To unsubscribe or change options:
https://lists.samba.org/mailman/listinfo/rsync
  >Before posting, read: http://www.catb.org/~esr/faqs/smart-questions.html


-- 
Ken Chase - ken at heavycomputing.ca skype:kenchase23 +1 416 897 6284 Toronto
Canada
Heavy Computing - Clued bandwidth, colocation and managed linux VPS @151 Front
St. W.