On Thu, Jan 20, 2005 at 11:51:42AM -0800, Peter Karman wrote:
> Anyone done any benchmarking lately of Swish-e vs Lucene (or Nutch)? I'm
> just curious...
I have done some benchmarks against Lucene and Clucene in past, and
swish is much faster (once I made note that Xapian is order of magnitude
slower than swish-e and got e-mail from Olly [Xapian developer] which
forced me to re-run tests and notice only double slowdown).
I am in process of benchmarking different search engines (it's not easy
because I have to take into account differences in query language and
special rules about what is indexed). For now, I have tested swish-e
(using old and new incremental file format), Xapian and mg (Managing
Gigabytes) and have following numbers to report:
real user sys
swish-e incremental 0.357 0.027 0.023
swish-e 0.327 0.057 0.036
xapian omega 0.458 0.029 0.016
mg 0.327 0.022 0.035
It's a search query for "full AND text AND search" on my test corpus of
316,817 unique words indexed.
62,544 files indexed. 691,244,737 total bytes. 55,650,106 total words.
Elapsed time: 00:11:32 CPU time: 00:05:03
as reported by swish-e. Tests where done 6 times, and before first
search memory caches where flush, so first search is much longer than
following ones (numbers above are averages).
I will write detailed report when I finish and post link to this list if
there is interest.
Dobrica Pavlinusic 2share!2flame email@example.com
Unix addict. Internet consultant. http://www.rot13.org/~dpavlin
Received on Thu Jan 20 12:40:39 2005