[
https://issues.apache.org/jira/browse/SOLR-122?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12468140
]
Coda Hale commented on SOLR-122:
--------------------------------
So I just ran two tests: indexing and querying. The first was indexing about
251K records (mostly short strings, floats, and integers), 1K at a time (the
server was stopped and the index deleted between these tests). The second was
querying ("id:[1 TO 1000]") and returning 100 rows. These tests were run on my
MacBook Pro while I checked my mail and looked at funny pictures of cats. Solr
was running on Jetty, with the config from the solrb directory, but a different
schema.
So... libxml-ruby is much faster:
user system total real
REXML indexing: 2107.480000 24.760000 2132.240000 (2421.141147)
libxml indexing: 728.810000 24.440000 753.250000 (1016.763167)
REXML querying: 25.520000 7.210000 32.730000 (873.637089)
libxml querying: 19.790000 6.020000 25.810000 (706.032189)
(For those unfamiliar with Ruby benchmarks, the first three numbers are
CPU-seconds, the last number in parentheses is the actual elapsed clock time in
seconds.)
Just to clarify, I'm not suggesting that REXML be ditched entirely; rather, I
think it would be better to have a drop-in replacement for people who want to
use libxml (or Xaagly or whatever) in situations where performance is important.
Also, my previous file had an escaping bug, so I'll post a corrected version.
(Basically, one should use XML::Node#<< instead of passing a second parameter
to XML::Node.new when setting the content of an element to an unescaped value.)
> Add optional support for Ruby-libxml2 (vs. REXML)
> -------------------------------------------------
>
> Key: SOLR-122
> URL: https://issues.apache.org/jira/browse/SOLR-122
> Project: Solr
> Issue Type: Improvement
> Components: clients - ruby - flare
> Reporter: Coda Hale
> Attachments: libxml.rb, libxml.rb
>
>
> This file adds drop-in support for the ruby-libxml2, which is a wrapper for
> the libxml2 library, which is an order of magnitude or so faster than REXML.
> This depends on my SOLR-121 patch for multi-document adds, since the behavior
> of Solr::Request::AddDocument#to_s is different.
> Requiring this makes some tests fail, but for trivial reasons: some tests are
> directly tied to REXML, others fail due to interelement whitespace added by
> libxml2 (which you can't disable via the Ruby interface). Functionally, it's
> identical, and passes all functional tests.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
